frame.py

"""
DataFrame
---------
An efficient 2D container for potentially mixed-type time series or other
labeled data series.

Similar to its R counterpart, data.frame, except providing automatic data
alignment and a host of useful data manipulation methods having to do with the
labeling information
"""
from __future__ import division
# pylint: disable=E1101,E1103
# pylint: disable=W0212,W0231,W0703,W0622

import functools
import collections
import itertools
import sys
import types
import warnings

from numpy import nan as NA
import numpy as np
import numpy.ma as ma

from pandas.core.common import (isnull, notnull, PandasError, _try_sort,
                                _default_index, _maybe_upcast, is_sequence,
                                _infer_dtype_from_scalar, _values_from_object,
                                is_list_like, _get_dtype, _maybe_box_datetimelike,
                                is_categorical_dtype, is_object_dtype, _possibly_infer_to_datetimelike)
from pandas.core.generic import NDFrame, _shared_docs
from pandas.core.index import Index, MultiIndex, _ensure_index
from pandas.core.indexing import (maybe_droplevels,
                                  convert_to_index_sliceable,
                                  check_bool_indexer)
from pandas.core.internals import (BlockManager,
                                   create_block_manager_from_arrays,
                                   create_block_manager_from_blocks)
from pandas.core.series import Series
from pandas.core.categorical import Categorical
import pandas.computation.expressions as expressions
from pandas.computation.eval import eval as _eval
from numpy import percentile as _quantile
from pandas.compat import(range, zip, lrange, lmap, lzip, StringIO, u,
                          OrderedDict, raise_with_traceback)
from pandas import compat
from pandas.sparse.array import SparseArray
from pandas.util.decorators import deprecate, Appender, Substitution, \
    deprecate_kwarg

from pandas.tseries.period import PeriodIndex
from pandas.tseries.index import DatetimeIndex

import pandas.core.algorithms as algos
import pandas.core.common as com
import pandas.core.format as fmt
import pandas.core.nanops as nanops
import pandas.core.ops as ops

import pandas.lib as lib
import pandas.algos as _algos

from pandas.core.config import get_option

#----------------------------------------------------------------------
# Docstring templates

_shared_doc_kwargs = dict(axes='index, columns', klass='DataFrame',
                          axes_single_arg="{0,1,'index','columns'}")

_numeric_only_doc = """numeric_only : boolean, default None
    Include only float, int, boolean data. If None, will attempt to use
    everything, then use only numeric data
"""

_merge_doc = """
Merge DataFrame objects by performing a database-style join operation by
columns or indexes.

If joining columns on columns, the DataFrame indexes *will be
ignored*. Otherwise if joining indexes on indexes or indexes on a column or
columns, the index will be passed on.

Parameters
----------%s
right : DataFrame
how : {'left', 'right', 'outer', 'inner'}, default 'inner'
    * left: use only keys from left frame (SQL: left outer join)
    * right: use only keys from right frame (SQL: right outer join)
    * outer: use union of keys from both frames (SQL: full outer join)
    * inner: use intersection of keys from both frames (SQL: inner join)
on : label or list
    Field names to join on. Must be found in both DataFrames. If on is
    None and not merging on indexes, then it merges on the intersection of
    the columns by default.
left_on : label or list, or array-like
    Field names to join on in left DataFrame. Can be a vector or list of
    vectors of the length of the DataFrame to use a particular vector as
    the join key instead of columns
right_on : label or list, or array-like
    Field names to join on in right DataFrame or vector/list of vectors per
    left_on docs
left_index : boolean, default False
    Use the index from the left DataFrame as the join key(s). If it is a
    MultiIndex, the number of keys in the other DataFrame (either the index
    or a number of columns) must match the number of levels
right_index : boolean, default False
    Use the index from the right DataFrame as the join key. Same caveats as
    left_index
sort : boolean, default False
    Sort the join keys lexicographically in the result DataFrame
suffixes : 2-length sequence (tuple, list, ...)
    Suffix to apply to overlapping column names in the left and right
    side, respectively
copy : boolean, default True
    If False, do not copy data unnecessarily

Examples
--------

>>> A              >>> B
    lkey value         rkey value
0   foo  1         0   foo  5
1   bar  2         1   bar  6
2   baz  3         2   qux  7
3   foo  4         3   bar  8

>>> merge(A, B, left_on='lkey', right_on='rkey', how='outer')
   lkey  value_x  rkey  value_y
0  foo   1        foo   5
1  foo   4        foo   5
2  bar   2        bar   6
3  bar   2        bar   8
4  baz   3        NaN   NaN
5  NaN   NaN      qux   7

Returns
-------
merged : DataFrame
    The output type will the be same as 'left', if it is a subclass
    of DataFrame.
"""

#----------------------------------------------------------------------
# DataFrame class


class DataFrame(NDFrame):

    """ Two-dimensional size-mutable, potentially heterogeneous tabular data
    structure with labeled axes (rows and columns). Arithmetic operations
    align on both row and column labels. Can be thought of as a dict-like
    container for Series objects. The primary pandas data structure

    Parameters
    ----------
    data : numpy ndarray (structured or homogeneous), dict, or DataFrame
        Dict can contain Series, arrays, constants, or list-like objects
    index : Index or array-like
        Index to use for resulting frame. Will default to np.arange(n) if
        no indexing information part of input data and no index provided
    columns : Index or array-like
        Column labels to use for resulting frame. Will default to
        np.arange(n) if no column labels are provided
    dtype : dtype, default None
        Data type to force, otherwise infer
    copy : boolean, default False
        Copy data from inputs. Only affects DataFrame / 2d ndarray input

    Examples
    --------
    >>> d = {'col1': ts1, 'col2': ts2}
    >>> df = DataFrame(data=d, index=index)
    >>> df2 = DataFrame(np.random.randn(10, 5))
    >>> df3 = DataFrame(np.random.randn(10, 5),
    ...                 columns=['a', 'b', 'c', 'd', 'e'])

    See also
    --------
    DataFrame.from_records : constructor from tuples, also record arrays
    DataFrame.from_dict : from dicts of Series, arrays, or dicts
    DataFrame.from_csv : from CSV files
    DataFrame.from_items : from sequence of (key, value) pairs
    pandas.read_csv, pandas.read_table, pandas.read_clipboard
    """
    _auto_consolidate = True

    @property
    def _constructor(self):
        return DataFrame

    _constructor_sliced = Series

    def __init__(self, data=None, index=None, columns=None, dtype=None,
                 copy=False):
        if data is None:
            data = {}
        if dtype is not None:
            dtype = self._validate_dtype(dtype)

        if isinstance(data, DataFrame):
            data = data._data

        if isinstance(data, BlockManager):
            mgr = self._init_mgr(data, axes=dict(index=index, columns=columns),
                                 dtype=dtype, copy=copy)
        elif isinstance(data, dict):
            mgr = self._init_dict(data, index, columns, dtype=dtype)
        elif isinstance(data, ma.MaskedArray):
            import numpy.ma.mrecords as mrecords
            # masked recarray
            if isinstance(data, mrecords.MaskedRecords):
                mgr = _masked_rec_array_to_mgr(data, index, columns, dtype,
                                               copy)

            # a masked array
            else:
                mask = ma.getmaskarray(data)
                if mask.any():
                    data, fill_value = _maybe_upcast(data, copy=True)
                    data[mask] = fill_value
                else:
                    data = data.copy()
                mgr = self._init_ndarray(data, index, columns, dtype=dtype,
                                         copy=copy)

        elif isinstance(data, (np.ndarray, Series, Index)):
            if data.dtype.names:
                data_columns = list(data.dtype.names)
                data = dict((k, data[k]) for k in data_columns)
                if columns is None:
                    columns = data_columns
                mgr = self._init_dict(data, index, columns, dtype=dtype)
            elif getattr(data, 'name', None):
                mgr = self._init_dict({data.name: data}, index, columns,
                                      dtype=dtype)
            else:
                mgr = self._init_ndarray(data, index, columns, dtype=dtype,
                                         copy=copy)
        elif isinstance(data, (list, types.GeneratorType)):
            if isinstance(data, types.GeneratorType):
                data = list(data)
            if len(data) > 0:
                if is_list_like(data[0]) and getattr(data[0], 'ndim', 1) == 1:
                    arrays, columns = _to_arrays(data, columns, dtype=dtype)
                    columns = _ensure_index(columns)

                    # set the index
                    if index is None:
                        if isinstance(data[0], Series):
                            index = _get_names_from_index(data)
                        elif isinstance(data[0], Categorical):
                            index = _default_index(len(data[0]))
                        else:
                            index = _default_index(len(data))

                    mgr = _arrays_to_mgr(arrays, columns, index, columns,
                                         dtype=dtype)
                else:
                    mgr = self._init_ndarray(data, index, columns, dtype=dtype,
                                             copy=copy)
            else:
                mgr = self._init_ndarray(data, index, columns, dtype=dtype,
                                         copy=copy)
        elif isinstance(data, collections.Iterator):
            raise TypeError("data argument can't be an iterator")
        else:
            try:
                arr = np.array(data, dtype=dtype, copy=copy)
            except (ValueError, TypeError) as e:
                exc = TypeError('DataFrame constructor called with '
                                'incompatible data and dtype: %s' % e)
                raise_with_traceback(exc)

            if arr.ndim == 0 and index is not None and columns is not None:
                if isinstance(data, compat.string_types) and dtype is None:
                    dtype = np.object_
                if dtype is None:
                    dtype, data = _infer_dtype_from_scalar(data)

                values = np.empty((len(index), len(columns)), dtype=dtype)
                values.fill(data)
                mgr = self._init_ndarray(values, index, columns, dtype=dtype,
                                         copy=False)
            else:
                raise PandasError('DataFrame constructor not properly called!')

        NDFrame.__init__(self, mgr, fastpath=True)

    def _init_dict(self, data, index, columns, dtype=None):
        """
        Segregate Series based on type and coerce into matrices.
        Needs to handle a lot of exceptional cases.
        """
        if columns is not None:
            columns = _ensure_index(columns)

            # prefilter if columns passed

            data = dict((k, v) for k, v in compat.iteritems(data)
                        if k in columns)

            if index is None:
                index = extract_index(list(data.values()))
            else:
                index = _ensure_index(index)

            arrays = []
            data_names = []
            for k in columns:
                if k not in data:
                    # no obvious "empty" int column
                    if dtype is not None and issubclass(dtype.type,
                                                        np.integer):
                        continue

                    if dtype is None:
                        # 1783
                        v = np.empty(len(index), dtype=object)
                    else:
                        v = np.empty(len(index), dtype=dtype)

                    v.fill(NA)
                else:
                    v = data[k]
                data_names.append(k)
                arrays.append(v)
        else:
            keys = list(data.keys())
            if not isinstance(data, OrderedDict):
                keys = _try_sort(keys)
            columns = data_names = Index(keys)
            arrays = [data[k] for k in keys]

        return _arrays_to_mgr(arrays, data_names, index, columns,
                              dtype=dtype)

    def _init_ndarray(self, values, index, columns, dtype=None,
                      copy=False):
        # input must be a ndarray, list, Series, index

        if isinstance(values, Series):
            if columns is None:
                if values.name is not None:
                    columns = [values.name]
            if index is None:
                index = values.index
            else:
                values = values.reindex(index)

            # zero len case (GH #2234)
            if not len(values) and columns is not None and len(columns):
                values = np.empty((0, 1), dtype=object)

        # helper to create the axes as indexes
        def _get_axes(N, K, index=index, columns=columns):
            # return axes or defaults

            if index is None:
                index = _default_index(N)
            else:
                index = _ensure_index(index)

            if columns is None:
                columns = _default_index(K)
            else:
                columns = _ensure_index(columns)
            return index, columns

        # we could have a categorical type passed or coerced to 'category'
        # recast this to an _arrays_to_mgr
        if is_categorical_dtype(getattr(values,'dtype',None)) or is_categorical_dtype(dtype):

            if not hasattr(values,'dtype'):
                values = _prep_ndarray(values, copy=copy)
                values = values.ravel()
            elif copy:
                values = values.copy()

            index, columns = _get_axes(len(values),1)
            return _arrays_to_mgr([ values ], columns, index, columns,
                                  dtype=dtype)

        # by definition an array here
        # the dtypes will be coerced to a single dtype
        values = _prep_ndarray(values, copy=copy)

        if dtype is not None:

            if values.dtype != dtype:
                try:
                    values = values.astype(dtype)
                except Exception as orig:
                    e = ValueError("failed to cast to '%s' (Exception was: %s)"
                                   % (dtype, orig))
                    raise_with_traceback(e)

        index, columns = _get_axes(*values.shape)
        values = values.T

        # if we don't have a dtype specified, then try to convert objects
        # on the entire block; this is to convert if we have datetimelike's
        # embedded in an object type
        if dtype is None and is_object_dtype(values):
            values = _possibly_infer_to_datetimelike(values)

        return create_block_manager_from_blocks([values], [columns, index])

    @property
    def axes(self):
        return [self.index, self.columns]

    @property
    def shape(self):
        return (len(self.index), len(self.columns))

    def _repr_fits_vertical_(self):
        """
        Check length against max_rows.
        """
        max_rows = get_option("display.max_rows")
        return len(self) <= max_rows

    def _repr_fits_horizontal_(self, ignore_width=False):
        """
        Check if full repr fits in horizontal boundaries imposed by the display
        options width and max_columns. In case off non-interactive session, no
        boundaries apply.

        ignore_width is here so ipnb+HTML output can behave the way
        users expect. display.max_columns remains in effect.
        GH3541, GH3573
        """

        width, height = fmt.get_console_size()
        max_columns = get_option("display.max_columns")
        nb_columns = len(self.columns)

        # exceed max columns
        if ((max_columns and nb_columns > max_columns) or
                ((not ignore_width) and width and nb_columns > (width // 2))):
            return False

        if (ignore_width  # used by repr_html under IPython notebook
                # scripts ignore terminal dims
                or not com.in_interactive_session()):
            return True

        if (get_option('display.width') is not None or
                com.in_ipython_frontend()):
            # check at least the column row for excessive width
            max_rows = 1
        else:
            max_rows = get_option("display.max_rows")

        # when auto-detecting, so width=None and not in ipython front end
        # check whether repr fits horizontal by actualy checking
        # the width of the rendered repr
        buf = StringIO()

        # only care about the stuff we'll actually print out
        # and to_string on entire frame may be expensive
        d = self

        if not (max_rows is None):  # unlimited rows
            # min of two, where one may be None
            d = d.iloc[:min(max_rows, len(d))]
        else:
            return True

        d.to_string(buf=buf)
        value = buf.getvalue()
        repr_width = max([len(l) for l in value.split('\n')])

        return repr_width < width

    def _info_repr(self):
        """True if the repr should show the info view."""
        info_repr_option = (get_option("display.large_repr") == "info")
        return info_repr_option and not (
            self._repr_fits_horizontal_() and self._repr_fits_vertical_()
        )

    def __unicode__(self):
        """
        Return a string representation for a particular DataFrame

        Invoked by unicode(df) in py2 only. Yields a Unicode String in both
        py2/py3.
        """
        buf = StringIO(u(""))
        if self._info_repr():
            self.info(buf=buf)
            return buf.getvalue()

        max_rows = get_option("display.max_rows")
        max_cols = get_option("display.max_columns")
        show_dimensions = get_option("display.show_dimensions")
        if get_option("display.expand_frame_repr"):
            width, _ = fmt.get_console_size()
        else:
            width = None
        self.to_string(buf=buf, max_rows=max_rows, max_cols=max_cols,
                       line_width=width, show_dimensions=show_dimensions)

        return buf.getvalue()

    def _repr_html_(self):
        """
        Return a html representation for a particular DataFrame.
        Mainly for IPython notebook.
        """
        # qtconsole doesn't report its line width, and also
        # behaves badly when outputting an HTML table
        # that doesn't fit the window, so disable it.
        # XXX: In IPython 3.x and above, the Qt console will not attempt to
        # display HTML, so this check can be removed when support for IPython 2.x
        # is no longer needed.
        if com.in_qtconsole():
            # 'HTML output is disabled in QtConsole'
            return None

        if self._info_repr():
            buf = StringIO(u(""))
            self.info(buf=buf)
            # need to escape the <class>, should be the first line.
            val = buf.getvalue().replace('<', r'&lt;', 1).replace('>',
                                                                  r'&gt;', 1)
            return '<pre>' + val + '</pre>'

        if get_option("display.notebook_repr_html"):
            max_rows = get_option("display.max_rows")
            max_cols = get_option("display.max_columns")
            show_dimensions = get_option("display.show_dimensions")

            return ('<div style="max-height:1000px;'
                    'max-width:1500px;overflow:auto;">\n' +
                    self.to_html(max_rows=max_rows, max_cols=max_cols,
                                 show_dimensions=show_dimensions) + '\n</div>')
        else:
            return None

    def iteritems(self):
        """Iterator over (column, series) pairs"""
        if self.columns.is_unique and hasattr(self, '_item_cache'):
            for k in self.columns:
                yield k, self._get_item_cache(k)
        else:
            for i, k in enumerate(self.columns):
                yield k, self.icol(i)

    def iterrows(self):
        """
        Iterate over rows of DataFrame as (index, Series) pairs.

        Notes
        -----

        * ``iterrows`` does **not** preserve dtypes across the rows (dtypes
          are preserved across columns for DataFrames). For example,

            >>> df = DataFrame([[1, 1.0]], columns=['x', 'y'])
            >>> row = next(df.iterrows())[1]
            >>> print(row['x'].dtype)
            float64
            >>> print(df['x'].dtype)
            int64

        Returns
        -------
        it : generator
            A generator that iterates over the rows of the frame.
        """
        columns = self.columns
        for k, v in zip(self.index, self.values):
            s = Series(v, index=columns, name=k)
            yield k, s

    def itertuples(self, index=True):
        """
        Iterate over rows of DataFrame as tuples, with index value
        as first element of the tuple
        """
        arrays = []
        if index:
            arrays.append(self.index)

        # use integer indexing because of possible duplicate column names
        arrays.extend(self.iloc[:, k] for k in range(len(self.columns)))
        return zip(*arrays)

    if compat.PY3:  # pragma: no cover
        items = iteritems

    def __len__(self):
        """Returns length of info axis, but here we use the index """
        return len(self.index)

    def dot(self, other):
        """
        Matrix multiplication with DataFrame or Series objects

        Parameters
        ----------
        other : DataFrame or Series

        Returns
        -------
        dot_product : DataFrame or Series
        """
        if isinstance(other, (Series, DataFrame)):
            common = self.columns.union(other.index)
            if (len(common) > len(self.columns) or
                    len(common) > len(other.index)):
                raise ValueError('matrices are not aligned')

            left = self.reindex(columns=common, copy=False)
            right = other.reindex(index=common, copy=False)
            lvals = left.values
            rvals = right.values
        else:
            left = self
            lvals = self.values
            rvals = np.asarray(other)
            if lvals.shape[1] != rvals.shape[0]:
                raise ValueError('Dot product shape mismatch, %s vs %s' %
                                 (lvals.shape, rvals.shape))

        if isinstance(other, DataFrame):
            return self._constructor(np.dot(lvals, rvals),
                                     index=left.index,
                                     columns=other.columns)
        elif isinstance(other, Series):
            return Series(np.dot(lvals, rvals), index=left.index)
        elif isinstance(rvals, (np.ndarray, Index)):
            result = np.dot(lvals, rvals)
            if result.ndim == 2:
                return self._constructor(result, index=left.index)
            else:
                return Series(result, index=left.index)
        else:  # pragma: no cover
            raise TypeError('unsupported type: %s' % type(other))

    #----------------------------------------------------------------------
    # IO methods (to / from other formats)

    @classmethod
    def from_dict(cls, data, orient='columns', dtype=None):
        """
        Construct DataFrame from dict of array-like or dicts

        Parameters
        ----------
        data : dict
            {field : array-like} or {field : dict}
        orient : {'columns', 'index'}, default 'columns'
            The "orientation" of the data. If the keys of the passed dict
            should be the columns of the resulting DataFrame, pass 'columns'
            (default). Otherwise if the keys should be rows, pass 'index'.

        Returns
        -------
        DataFrame
        """
        index, columns = None, None
        orient = orient.lower()
        if orient == 'index':
            if len(data) > 0:
                # TODO speed up Series case
                if isinstance(list(data.values())[0], (Series, dict)):
                    data = _from_nested_dict(data)
                else:
                    data, index = list(data.values()), list(data.keys())
        elif orient != 'columns':  # pragma: no cover
            raise ValueError('only recognize index or columns for orient')

        return cls(data, index=index, columns=columns, dtype=dtype)

    @deprecate_kwarg(old_arg_name='outtype', new_arg_name='orient')
    def to_dict(self, orient='dict'):
        """Convert DataFrame to dictionary.

        Parameters
        ----------
        orient : str {'dict', 'list', 'series', 'split', 'records'}
            Determines the type of the values of the dictionary.

            - dict (default) : dict like {column -> {index -> value}}
            - list : dict like {column -> [values]}
            - series : dict like {column -> Series(values)}
            - split : dict like
              {index -> [index], columns -> [columns], data -> [values]}
            - records : list like
              [{column -> value}, ... , {column -> value}]

            Abbreviations are allowed. `s` indicates `series` and `sp`
            indicates `split`.

        Returns
        -------
        result : dict like {column -> {index -> value}}
        """
        if not self.columns.is_unique:
            warnings.warn("DataFrame columns are not unique, some "
                          "columns will be omitted.", UserWarning)
        if orient.lower().startswith('d'):
            return dict((k, v.to_dict()) for k, v in compat.iteritems(self))
        elif orient.lower().startswith('l'):
            return dict((k, v.tolist()) for k, v in compat.iteritems(self))
        elif orient.lower().startswith('sp'):
            return {'index': self.index.tolist(),
                    'columns': self.columns.tolist(),
                    'data': self.values.tolist()}
        elif orient.lower().startswith('s'):
            return dict((k, v) for k, v in compat.iteritems(self))
        elif orient.lower().startswith('r'):
            return [dict((k, v) for k, v in zip(self.columns, row))
                    for row in self.values]
        else:
            raise ValueError("orient '%s' not understood" % orient)

    def to_gbq(self, destination_table, project_id=None, chunksize=10000,
               verbose=True, reauth=False):
        """Write a DataFrame to a Google BigQuery table.

        THIS IS AN EXPERIMENTAL LIBRARY

        If the table exists, the dataframe will be written to the table using
        the defined table schema and column types. For simplicity, this method
        uses the Google BigQuery streaming API. The to_gbq method chunks data
        into a default chunk size of 10,000. Failures return the complete error
        response which can be quite long depending on the size of the insert.
        There are several important limitations of the Google streaming API
        which are detailed at:
        https://developers.google.com/bigquery/streaming-data-into-bigquery.

        Parameters
        ----------
        dataframe : DataFrame
            DataFrame to be written
        destination_table : string
            Name of table to be written, in the form 'dataset.tablename'
        project_id : str
            Google BigQuery Account project ID.
        chunksize : int (default 10000)
            Number of rows to be inserted in each chunk from the dataframe.
        verbose : boolean (default True)
            Show percentage complete
        reauth : boolean (default False)
            Force Google BigQuery to reauthenticate the user. This is useful
            if multiple accounts are used.

        """

        from pandas.io import gbq
        return gbq.to_gbq(self, destination_table, project_id=project_id,
                          chunksize=chunksize, verbose=verbose,
                          reauth=reauth)

    @classmethod
    def from_records(cls, data, index=None, exclude=None, columns=None,
                     coerce_float=False, nrows=None):
        """
        Convert structured or record ndarray to DataFrame

        Parameters
        ----------
        data : ndarray (structured dtype), list of tuples, dict, or DataFrame
        index : string, list of fields, array-like
            Field of array to use as the index, alternately a specific set of
            input labels to use
        exclude : sequence, default None
            Columns or fields to exclude
        columns : sequence, default None
            Column names to use. If the passed data do not have names
            associated with them, this argument provides names for the
            columns. Otherwise this argument indicates the order of the columns
            in the result (any names not found in the data will become all-NA
            columns)
        coerce_float : boolean, default False
            Attempt to convert values to non-string, non-numeric objects (like
            decimal.Decimal) to floating point, useful for SQL result sets

        Returns
        -------
        df : DataFrame
        """
        # Make a copy of the input columns so we can modify it
        if columns is not None:
            columns = _ensure_index(columns)

        if com.is_iterator(data):
            if nrows == 0:
                return cls()

            try:
                if compat.PY3:
                    first_row = next(data)
                else:
                    first_row = next(data)
            except StopIteration:
                return cls(index=index, columns=columns)

            dtype = None
            if hasattr(first_row, 'dtype') and first_row.dtype.names:
                dtype = first_row.dtype

            values = [first_row]

            if nrows is None:
                values += data
            else:
                values.extend(itertools.islice(data, nrows - 1))

            if dtype is not None:
                data = np.array(values, dtype=dtype)
            else:
                data = values

        if isinstance(data, dict):
            if columns is None:
                columns = arr_columns = _ensure_index(sorted(data))
                arrays = [data[k] for k in columns]
            else:
                arrays = []
                arr_columns = []
                for k, v in compat.iteritems(data):
                    if k in columns:
                        arr_columns.append(k)
                        arrays.append(v)

                arrays, arr_columns = _reorder_arrays(arrays, arr_columns,
                                                      columns)

        elif isinstance(data, (np.ndarray, DataFrame)):
            arrays, columns = _to_arrays(data, columns)
            if columns is not None:
                columns = _ensure_index(columns)
            arr_columns = columns
        else:
            arrays, arr_columns = _to_arrays(data, columns,
                                             coerce_float=coerce_float)

            arr_columns = _ensure_index(arr_columns)
            if columns is not None:
                columns = _ensure_index(columns)
            else:
                columns = arr_columns

        if exclude is None:
            exclude = set()
        else:
            exclude = set(exclude)

        result_index = None
        if index is not None:
            if (isinstance(index, compat.string_types) or
                    not hasattr(index, "__iter__")):
                i = columns.get_loc(index)
                exclude.add(index)
                if len(arrays) > 0:
                    result_index = Index(arrays[i], name=index)
                else:
                    result_index = Index([], name=index)
            else:
                try:
                    to_remove = [arr_columns.get_loc(field) for field in index]

                    result_index = MultiIndex.from_arrays(
                        [arrays[i] for i in to_remove], names=index)

                    exclude.update(index)
                except Exception:
                    result_index = index

        if any(exclude):
            arr_exclude = [x for x in exclude if x in arr_columns]
            to_remove = [arr_columns.get_loc(col) for col in arr_exclude]
            arrays = [v for i, v in enumerate(arrays) if i not in to_remove]

            arr_columns = arr_columns.drop(arr_exclude)
            columns = columns.drop(exclude)

        mgr = _arrays_to_mgr(arrays, arr_columns, result_index,
                             columns)

        return cls(mgr)

    def to_records(self, index=True, convert_datetime64=True):
        """
        Convert DataFrame to record array. Index will be put in the
        'index' field of the record array if requested

        Parameters
        ----------
        index : boolean, default True
            Include index in resulting record array, stored in 'index' field
        convert_datetime64 : boolean, default True
            Whether to convert the index to datetime.datetime if it is a
            DatetimeIndex

        Returns
        -------
        y : recarray
        """
        if index:
            if com.is_datetime64_dtype(self.index) and convert_datetime64:
                ix_vals = [self.index.to_pydatetime()]
            else:
                if isinstance(self.index, MultiIndex):
                    # array of tuples to numpy cols. copy copy copy
                    ix_vals = lmap(np.array, zip(*self.index.values))
                else:
                    ix_vals = [self.index.values]

            arrays = ix_vals + [self[c].get_values() for c in self.columns]

            count = 0
            index_names = list(self.index.names)
            if isinstance(self.index, MultiIndex):
                for i, n in enumerate(index_names):
                    if n is None:
                        index_names[i] = 'level_%d' % count
                        count += 1
            elif index_names[0] is None:
                index_names = ['index']
            names = index_names + lmap(str, self.columns)
        else:
            arrays = [self[c].get_values() for c in self.columns]
            names = lmap(str, self.columns)

        dtype = np.dtype([(x, v.dtype) for x, v in zip(names, arrays)])
        return np.rec.fromarrays(arrays, dtype=dtype, names=names)

    @classmethod
    def from_items(cls, items, columns=None, orient='columns'):
        """
        Convert (key, value) pairs to DataFrame. The keys will be the axis
        index (usually the columns, but depends on the specified
        orientation). The values should be arrays or Series.

        Parameters
        ----------
        items : sequence of (key, value) pairs
            Values should be arrays or Series.
        columns : sequence of column labels, optional
            Must be passed if orient='index'.
        orient : {'columns', 'index'}, default 'columns'
            The "orientation" of the data. If the keys of the
            input correspond to column labels, pass 'columns'
            (default). Otherwise if the keys correspond to the index,
            pass 'index'.

        Returns
        -------
        frame : DataFrame
        """
        keys, values = lzip(*items)

        if orient == 'columns':
            if columns is not None:
                columns = _ensure_index(columns)

                idict = dict(items)
                if len(idict) < len(items):
                    if not columns.equals(_ensure_index(keys)):
                        raise ValueError('With non-unique item names, passed '
                                         'columns must be identical')
                    arrays = values
                else:
                    arrays = [idict[k] for k in columns if k in idict]
            else:
                columns = _ensure_index(keys)
                arrays = values

            return cls._from_arrays(arrays, columns, None)
        elif orient == 'index':
            if columns is None:
                raise TypeError("Must pass columns with orient='index'")

            keys = _ensure_index(keys)

            arr = np.array(values, dtype=object).T
            data = [lib.maybe_convert_objects(v) for v in arr]
            return cls._from_arrays(data, columns, keys)
        else:  # pragma: no cover
            raise ValueError("'orient' must be either 'columns' or 'index'")

    @classmethod
    def _from_arrays(cls, arrays, columns, index, dtype=None):
        mgr = _arrays_to_mgr(arrays, columns, index, columns, dtype=dtype)
        return cls(mgr)

    @classmethod
    def from_csv(cls, path, header=0, sep=',', index_col=0,
                 parse_dates=True, encoding=None, tupleize_cols=False,
                 infer_datetime_format=False):
        """
        Read delimited file into DataFrame

        Parameters
        ----------
        path : string file path or file handle / StringIO
        header : int, default 0
            Row to use at header (skip prior rows)
        sep : string, default ','
            Field delimiter
        index_col : int or sequence, default 0
            Column to use for index. If a sequence is given, a MultiIndex
            is used. Different default from read_table
        parse_dates : boolean, default True
            Parse dates. Different default from read_table
        tupleize_cols : boolean, default False
            write multi_index columns as a list of tuples (if True)
            or new (expanded format) if False)
        infer_datetime_format: boolean, default False
            If True and `parse_dates` is True for a column, try to infer the
            datetime format based on the first datetime string. If the format
            can be inferred, there often will be a large parsing speed-up.

        Notes
        -----
        Preferable to use read_table for most general purposes but from_csv
        makes for an easy roundtrip to and from file, especially with a
        DataFrame of time series data

        Returns
        -------
        y : DataFrame
        """
        from pandas.io.parsers import read_table
        return read_table(path, header=header, sep=sep,
                          parse_dates=parse_dates, index_col=index_col,
                          encoding=encoding, tupleize_cols=tupleize_cols,
                          infer_datetime_format=infer_datetime_format)

    def to_sparse(self, fill_value=None, kind='block'):
        """
        Convert to SparseDataFrame

        Parameters
        ----------
        fill_value : float, default NaN
        kind : {'block', 'integer'}

        Returns
        -------
        y : SparseDataFrame
        """
        from pandas.core.sparse import SparseDataFrame
        return SparseDataFrame(self._series, index=self.index,
                               default_kind=kind,
                               default_fill_value=fill_value)

    def to_panel(self):
        """
        Transform long (stacked) format (DataFrame) into wide (3D, Panel)
        format.

        Currently the index of the DataFrame must be a 2-level MultiIndex. This
        may be generalized later

        Returns
        -------
        panel : Panel
        """
        from pandas.core.panel import Panel

        # only support this kind for now
        if (not isinstance(self.index, MultiIndex) or  # pragma: no cover
                len(self.index.levels) != 2):
            raise NotImplementedError('Only 2-level MultiIndex are supported.')

        if not self.index.is_unique:
            raise ValueError("Can't convert non-uniquely indexed "
                             "DataFrame to Panel")

        self._consolidate_inplace()

        # minor axis must be sorted
        if self.index.lexsort_depth < 2:
            selfsorted = self.sortlevel(0)
        else:
            selfsorted = self

        major_axis, minor_axis = selfsorted.index.levels
        major_labels, minor_labels = selfsorted.index.labels
        shape = len(major_axis), len(minor_axis)

        # preserve names, if any
        major_axis = major_axis.copy()
        major_axis.name = self.index.names[0]

        minor_axis = minor_axis.copy()
        minor_axis.name = self.index.names[1]

        # create new axes
        new_axes = [selfsorted.columns, major_axis, minor_axis]

        # create new manager
        new_mgr = selfsorted._data.reshape_nd(axes=new_axes,
                                              labels=[major_labels, minor_labels],
                                              shape=shape,
                                              ref_items=selfsorted.columns)

        return Panel(new_mgr)

    to_wide = deprecate('to_wide', to_panel)

    @deprecate_kwarg(old_arg_name='cols', new_arg_name='columns')
    def to_csv(self, path_or_buf=None, sep=",", na_rep='', float_format=None,
               columns=None, header=True, index=True, index_label=None,
               mode='w', encoding=None, quoting=None,
               quotechar='"', line_terminator='\n', chunksize=None,
               tupleize_cols=False, date_format=None, doublequote=True,
               escapechar=None, **kwds):
        r"""Write DataFrame to a comma-separated values (csv) file

        Parameters
        ----------
        path_or_buf : string or file handle, default None
            File path or object, if None is provided the result is returned as
            a string.
        sep : character, default ","
            Field delimiter for the output file.
        na_rep : string, default ''
            Missing data representation
        float_format : string, default None
            Format string for floating point numbers
        columns : sequence, optional
            Columns to write
        header : boolean or list of string, default True
            Write out column names. If a list of string is given it is assumed
            to be aliases for the column names
        index : boolean, default True
            Write row names (index)
        index_label : string or sequence, or False, default None
            Column label for index column(s) if desired. If None is given, and
            `header` and `index` are True, then the index names are used. A
            sequence should be given if the DataFrame uses MultiIndex.  If
            False do not print fields for index names. Use index_label=False
            for easier importing in R
        nanRep : None
            deprecated, use na_rep
        mode : str
            Python write mode, default 'w'
        encoding : string, optional
            A string representing the encoding to use in the output file,
            defaults to 'ascii' on Python 2 and 'utf-8' on Python 3.
        line_terminator : string, default '\\n'
            The newline character or character sequence to use in the output
            file
        quoting : optional constant from csv module
            defaults to csv.QUOTE_MINIMAL
        quotechar : string (length 1), default '"'
            character used to quote fields
        doublequote : boolean, default True
            Control quoting of `quotechar` inside a field
        escapechar : string (length 1), default None
            character used to escape `sep` and `quotechar` when appropriate
        chunksize : int or None
            rows to write at a time
        tupleize_cols : boolean, default False
            write multi_index columns as a list of tuples (if True)
            or new (expanded format) if False)
        date_format : string, default None
            Format string for datetime objects
        cols : kwarg only alias of columns [deprecated]
        """

        formatter = fmt.CSVFormatter(self, path_or_buf,
                                     line_terminator=line_terminator,
                                     sep=sep, encoding=encoding,
                                     quoting=quoting, na_rep=na_rep,
                                     float_format=float_format, cols=columns,
                                     header=header, index=index,
                                     index_label=index_label, mode=mode,
                                     chunksize=chunksize, quotechar=quotechar,
                                     engine=kwds.get("engine"),
                                     tupleize_cols=tupleize_cols,
                                     date_format=date_format,
                                     doublequote=doublequote,
                                     escapechar=escapechar)
        formatter.save()

        if path_or_buf is None:
            return formatter.path_or_buf.getvalue()

    @deprecate_kwarg(old_arg_name='cols', new_arg_name='columns')
    def to_excel(self, excel_writer, sheet_name='Sheet1', na_rep='',
                 float_format=None, columns=None, header=True, index=True,
                 index_label=None, startrow=0, startcol=0, engine=None,
                 merge_cells=True, encoding=None, inf_rep='inf'):
        """
        Write DataFrame to a excel sheet

        Parameters
        ----------
        excel_writer : string or ExcelWriter object
            File path or existing ExcelWriter
        sheet_name : string, default 'Sheet1'
            Name of sheet which will contain DataFrame
        na_rep : string, default ''
            Missing data representation
        float_format : string, default None
            Format string for floating point numbers
        columns : sequence, optional
            Columns to write
        header : boolean or list of string, default True
            Write out column names. If a list of string is given it is
            assumed to be aliases for the column names
        index : boolean, default True
            Write row names (index)
        index_label : string or sequence, default None
            Column label for index column(s) if desired. If None is given, and
            `header` and `index` are True, then the index names are used. A
            sequence should be given if the DataFrame uses MultiIndex.
        startrow :
            upper left cell row to dump data frame
        startcol :
            upper left cell column to dump data frame
        engine : string, default None
            write engine to use - you can also set this via the options
            ``io.excel.xlsx.writer``, ``io.excel.xls.writer``, and
            ``io.excel.xlsm.writer``.
        merge_cells : boolean, default True
            Write MultiIndex and Hierarchical Rows as merged cells.
        encoding: string, default None
            encoding of the resulting excel file. Only necessary for xlwt,
            other writers support unicode natively.
        cols : kwarg only alias of columns [deprecated]
        inf_rep : string, default 'inf'
            Representation for infinity (there is no native representation for
            infinity in Excel)

        Notes
        -----
        If passing an existing ExcelWriter object, then the sheet will be added
        to the existing workbook.  This can be used to save different
        DataFrames to one workbook:

        >>> writer = ExcelWriter('output.xlsx')
        >>> df1.to_excel(writer,'Sheet1')
        >>> df2.to_excel(writer,'Sheet2')
        >>> writer.save()
        """
        from pandas.io.excel import ExcelWriter

        need_save = False
        if encoding == None:
            encoding = 'ascii'

        if isinstance(excel_writer, compat.string_types):
            excel_writer = ExcelWriter(excel_writer, engine=engine)
            need_save = True

        formatter = fmt.ExcelFormatter(self,
                                       na_rep=na_rep,
                                       cols=columns,
                                       header=header,
                                       float_format=float_format,
                                       index=index,
                                       index_label=index_label,
                                       merge_cells=merge_cells,
                                       inf_rep=inf_rep)
        formatted_cells = formatter.get_formatted_cells()
        excel_writer.write_cells(formatted_cells, sheet_name,
                                 startrow=startrow, startcol=startcol)
        if need_save:
            excel_writer.save()

    def to_stata(
        self, fname, convert_dates=None, write_index=True, encoding="latin-1",
            byteorder=None, time_stamp=None, data_label=None):
        """
        A class for writing Stata binary dta files from array-like objects

        Parameters
        ----------
        fname : file path or buffer
            Where to save the dta file.
        convert_dates : dict
            Dictionary mapping column of datetime types to the stata internal
            format that you want to use for the dates. Options are
            'tc', 'td', 'tm', 'tw', 'th', 'tq', 'ty'. Column can be either a
            number or a name.
        encoding : str
            Default is latin-1. Note that Stata does not support unicode.
        byteorder : str
            Can be ">", "<", "little", or "big". The default is None which uses
            `sys.byteorder`

        Examples
        --------
        >>> writer = StataWriter('./data_file.dta', data)
        >>> writer.write_file()

        Or with dates

        >>> writer = StataWriter('./date_data_file.dta', data, {2 : 'tw'})
        >>> writer.write_file()
        """
        from pandas.io.stata import StataWriter
        writer = StataWriter(fname, self, convert_dates=convert_dates,
                             encoding=encoding, byteorder=byteorder,
                             time_stamp=time_stamp, data_label=data_label,
                             write_index=write_index)
        writer.write_file()

    @Appender(fmt.docstring_to_string, indents=1)
    def to_string(self, buf=None, columns=None, col_space=None, colSpace=None,
                  header=True, index=True, na_rep='NaN', formatters=None,
                  float_format=None, sparsify=None, index_names=True,
                  justify=None, line_width=None, max_rows=None, max_cols=None,
                  show_dimensions=False):
        """
        Render a DataFrame to a console-friendly tabular output.
        """

        if colSpace is not None:  # pragma: no cover
            warnings.warn("colSpace is deprecated, use col_space",
                          FutureWarning)
            col_space = colSpace

        formatter = fmt.DataFrameFormatter(self, buf=buf, columns=columns,
                                           col_space=col_space, na_rep=na_rep,
                                           formatters=formatters,
                                           float_format=float_format,
                                           sparsify=sparsify,
                                           justify=justify,
                                           index_names=index_names,
                                           header=header, index=index,
                                           line_width=line_width,
                                           max_rows=max_rows,
                                           max_cols=max_cols,
                                           show_dimensions=show_dimensions)
        formatter.to_string()

        if buf is None:
            result = formatter.buf.getvalue()
            return result

    @Appender(fmt.docstring_to_string, indents=1)
    def to_html(self, buf=None, columns=None, col_space=None, colSpace=None,
                header=True, index=True, na_rep='NaN', formatters=None,
                float_format=None, sparsify=None, index_names=True,
                justify=None, bold_rows=True, classes=None, escape=True,
                max_rows=None, max_cols=None, show_dimensions=False):
        """
        Render a DataFrame as an HTML table.

        `to_html`-specific options:

        bold_rows : boolean, default True
            Make the row labels bold in the output
        classes : str or list or tuple, default None
            CSS class(es) to apply to the resulting html table
        escape : boolean, default True
            Convert the characters <, >, and & to HTML-safe sequences.=
        max_rows : int, optional
            Maximum number of rows to show before truncating. If None, show
            all.
        max_cols : int, optional
            Maximum number of columns to show before truncating. If None, show
            all.

        """

        if colSpace is not None:  # pragma: no cover
            warnings.warn("colSpace is deprecated, use col_space",
                          FutureWarning)
            col_space = colSpace

        formatter = fmt.DataFrameFormatter(self, buf=buf, columns=columns,
                                           col_space=col_space, na_rep=na_rep,
                                           formatters=formatters,
                                           float_format=float_format,
                                           sparsify=sparsify,
                                           justify=justify,
                                           index_names=index_names,
                                           header=header, index=index,
                                           bold_rows=bold_rows,
                                           escape=escape,
                                           max_rows=max_rows,
                                           max_cols=max_cols,
                                           show_dimensions=show_dimensions)
        formatter.to_html(classes=classes)

        if buf is None:
            return formatter.buf.getvalue()

    @Appender(fmt.docstring_to_string, indents=1)
    def to_latex(self, buf=None, columns=None, col_space=None, colSpace=None,
                 header=True, index=True, na_rep='NaN', formatters=None,
                 float_format=None, sparsify=None, index_names=True,
                 bold_rows=True, longtable=False, escape=True):
        """
        Render a DataFrame to a tabular environment table. You can splice
        this into a LaTeX document. Requires \\usepackage{booktabs}.

        `to_latex`-specific options:

        bold_rows : boolean, default True
            Make the row labels bold in the output
        longtable : boolean, default False
            Use a longtable environment instead of tabular. Requires adding
            a \\usepackage{longtable} to your LaTeX preamble.
        escape : boolean, default True
            When set to False prevents from escaping latex special
            characters in column names.

        """

        if colSpace is not None:  # pragma: no cover
            warnings.warn("colSpace is deprecated, use col_space",
                          FutureWarning)
            col_space = colSpace

        formatter = fmt.DataFrameFormatter(self, buf=buf, columns=columns,
                                           col_space=col_space, na_rep=na_rep,
                                           header=header, index=index,
                                           formatters=formatters,
                                           float_format=float_format,
                                           bold_rows=bold_rows,
                                           sparsify=sparsify,
                                           index_names=index_names,
                                           escape=escape)
        formatter.to_latex(longtable=longtable)

        if buf is None:
            return formatter.buf.getvalue()

    def info(self, verbose=None, buf=None, max_cols=None, memory_usage=None, null_counts=None):
        """
        Concise summary of a DataFrame.

        Parameters
        ----------
        verbose : {None, True, False}, optional
            Whether to print the full summary.
            None follows the `display.max_info_columns` setting.
            True or False overrides the `display.max_info_columns` setting.
        buf : writable buffer, defaults to sys.stdout
        max_cols : int, default None
            Determines whether full summary or short summary is printed.
            None follows the `display.max_info_columns` setting.
        memory_usage : boolean, default None
            Specifies whether total memory usage of the DataFrame
            elements (including index) should be displayed. None follows
            the `display.memory_usage` setting. True or False overrides
            the `display.memory_usage` setting. Memory usage is shown in
            human-readable units (base-2 representation).
        null_counts : boolean, default None
            Whether to show the non-null counts
            If None, then only show if the frame is smaller than max_info_rows and max_info_columns.
            If True, always show counts.
            If False, never show counts.

        """
        from pandas.core.format import _put_lines

        if buf is None:  # pragma: no cover
            buf = sys.stdout

        lines = []

        lines.append(str(type(self)))
        lines.append(self.index.summary())

        if len(self.columns) == 0:
            lines.append('Empty %s' % type(self).__name__)
            _put_lines(buf, lines)
            return

        cols = self.columns

        # hack
        if max_cols is None:
            max_cols = get_option(
                'display.max_info_columns', len(self.columns) + 1)

        max_rows = get_option('display.max_info_rows', len(self) + 1)

        if null_counts is None:
            show_counts = ((len(self.columns) <= max_cols) and
                           (len(self) < max_rows))
        else:
            show_counts = null_counts
        exceeds_info_cols = len(self.columns) > max_cols

        def _verbose_repr():
            lines.append('Data columns (total %d columns):' %
                         len(self.columns))
            space = max([len(com.pprint_thing(k)) for k in self.columns]) + 4
            counts = None

            tmpl = "%s%s"
            if show_counts:
                counts = self.count()
                if len(cols) != len(counts):  # pragma: no cover
                    raise AssertionError('Columns must equal counts (%d != %d)' %
                                         (len(cols), len(counts)))
                tmpl = "%s non-null %s"

            dtypes = self.dtypes
            for i, col in enumerate(self.columns):
                dtype = dtypes[col]
                col = com.pprint_thing(col)

                count = ""
                if show_counts:
                    count = counts.iloc[i]

                lines.append(_put_str(col, space) +
                             tmpl % (count, dtype))

        def _non_verbose_repr():
            lines.append(self.columns.summary(name='Columns'))

        def _sizeof_fmt(num, size_qualifier):
            # returns size in human readable format
            for x in ['bytes', 'KB', 'MB', 'GB', 'TB']:
                if num < 1024.0:
                    return "%3.1f%s %s" % (num, size_qualifier, x)
                num /= 1024.0
            return "%3.1f%s %s" % (num, size_qualifier, 'PB')

        if verbose:
            _verbose_repr()
        elif verbose is False:  # specifically set to False, not nesc None
            _non_verbose_repr()
        else:
            if exceeds_info_cols:
                _non_verbose_repr()
            else:
                _verbose_repr()

        counts = self.get_dtype_counts()
        dtypes = ['%s(%d)' % k for k in sorted(compat.iteritems(counts))]
        lines.append('dtypes: %s' % ', '.join(dtypes))
        if memory_usage is None:
            memory_usage = get_option('display.memory_usage')
        if memory_usage:  # append memory usage of df to display
            # size_qualifier is just a best effort; not guaranteed to catch all
            # cases (e.g., it misses categorical data even with object
            # categories)
            size_qualifier = ('+' if 'object' in counts
                              or is_object_dtype(self.index) else '')
            mem_usage = self.memory_usage(index=True).sum()
            lines.append("memory usage: %s\n" %
                            _sizeof_fmt(mem_usage, size_qualifier))
        _put_lines(buf, lines)

    def memory_usage(self, index=False):
        """Memory usage of DataFrame columns.

        Parameters
        ----------
        index : bool
            Specifies whether to include memory usage of DataFrame's
            index in returned Series. If `index=True` (default is False)
            the first index of the Series is `Index`.

        Returns
        -------
        sizes : Series
            A series with column names as index and memory usage of
            columns with units of bytes.

        Notes
        -----
        Memory usage does not include memory consumed by elements that
        are not components of the array.

        See Also
        --------
        numpy.ndarray.nbytes
        """
        result = Series([ c.values.nbytes for col, c in self.iteritems() ],
                        index=self.columns)
        if index:
             result = Series(self.index.nbytes,
                        index=['Index']).append(result)
        return result

    def transpose(self):
        """Transpose index and columns"""
        return super(DataFrame, self).transpose(1, 0)

    T = property(transpose)

    #----------------------------------------------------------------------
    # Picklability

    # legacy pickle formats
    def _unpickle_frame_compat(self, state):  # pragma: no cover
        from pandas.core.common import _unpickle_array
        if len(state) == 2:  # pragma: no cover
            series, idx = state
            columns = sorted(series)
        else:
            series, cols, idx = state
            columns = _unpickle_array(cols)

        index = _unpickle_array(idx)
        self._data = self._init_dict(series, index, columns, None)

    def _unpickle_matrix_compat(self, state):  # pragma: no cover
        from pandas.core.common import _unpickle_array
        # old unpickling
        (vals, idx, cols), object_state = state

        index = _unpickle_array(idx)
        dm = DataFrame(vals, index=index, columns=_unpickle_array(cols),
                       copy=False)

        if object_state is not None:
            ovals, _, ocols = object_state
            objects = DataFrame(ovals, index=index,
                                columns=_unpickle_array(ocols),
                                copy=False)

            dm = dm.join(objects)

        self._data = dm._data

    #----------------------------------------------------------------------
    #----------------------------------------------------------------------
    # Getting and setting elements

    def get_value(self, index, col, takeable=False):
        """
        Quickly retrieve single value at passed column and index

        Parameters
        ----------
        index : row label
        col : column label
        takeable : interpret the index/col as indexers, default False

        Returns
        -------
        value : scalar value
        """

        if takeable:
            series = self._iget_item_cache(col)
            return _maybe_box_datetimelike(series.values[index])

        series = self._get_item_cache(col)
        engine = self.index._engine
        return engine.get_value(series.get_values(), index)

    def set_value(self, index, col, value, takeable=False):
        """
        Put single value at passed column and index

        Parameters
        ----------
        index : row label
        col : column label
        value : scalar value
        takeable : interpret the index/col as indexers, default False

        Returns
        -------
        frame : DataFrame
            If label pair is contained, will be reference to calling DataFrame,
            otherwise a new object
        """
        try:
            if takeable is True:
                series = self._iget_item_cache(col)
                return series.set_value(index, value, takeable=True)

            series = self._get_item_cache(col)
            engine = self.index._engine
            engine.set_value(series.values, index, value)
            return self
        except (KeyError, TypeError):

            # set using a non-recursive method & reset the cache
            self.loc[index, col] = value
            self._item_cache.pop(col, None)

            return self

    def irow(self, i, copy=False):
        return self._ixs(i, axis=0)

    def icol(self, i):
        return self._ixs(i, axis=1)

    def _ixs(self, i, axis=0):
        """
        i : int, slice, or sequence of integers
        axis : int
        """

        # irow
        if axis == 0:

            """
            Notes
            -----
            If slice passed, the resulting data will be a view
            """

            if isinstance(i, slice):
                return self[i]
            else:
                label = self.index[i]
                if isinstance(label, Index):
                    # a location index by definition
                    result = self.take(i, axis=axis)
                    copy=True
                else:
                    new_values = self._data.fast_xs(i)

                    # if we are a copy, mark as such
                    copy = isinstance(new_values,np.ndarray) and new_values.base is None
                    result = Series(new_values, index=self.columns,
                                    name=self.index[i], dtype=new_values.dtype)
                result._set_is_copy(self, copy=copy)
                return result

        # icol
        else:

            """
            Notes
            -----
            If slice passed, the resulting data will be a view
            """

            label = self.columns[i]
            if isinstance(i, slice):
                # need to return view
                lab_slice = slice(label[0], label[-1])
                return self.ix[:, lab_slice]
            else:
                label = self.columns[i]
                if isinstance(label, Index):
                    return self.take(i, axis=1, convert=True)

                # if the values returned are not the same length
                # as the index (iow a not found value), iget returns
                # a 0-len ndarray. This is effectively catching
                # a numpy error (as numpy should really raise)
                values = self._data.iget(i)
                if not len(values):
                    values = np.array([np.nan] * len(self.index), dtype=object)
                result = self._constructor_sliced.from_array(
                    values, index=self.index,
                    name=label, fastpath=True)

                # this is a cached value, mark it so
                result._set_as_cached(label, self)

                return result

    def iget_value(self, i, j):
        return self.iat[i, j]

    def __getitem__(self, key):

        # shortcut if we are an actual column
        is_mi_columns = isinstance(self.columns, MultiIndex)
        try:
            if key in self.columns and not is_mi_columns:
                return self._getitem_column(key)
        except:
            pass

        # see if we can slice the rows
        indexer = convert_to_index_sliceable(self, key)
        if indexer is not None:
            return self._getitem_slice(indexer)

        if isinstance(key, (Series, np.ndarray, Index, list)):
            # either boolean or fancy integer index
            return self._getitem_array(key)
        elif isinstance(key, DataFrame):
            return self._getitem_frame(key)
        elif is_mi_columns:
            return self._getitem_multilevel(key)
        else:
            return self._getitem_column(key)

    def _getitem_column(self, key):
        """ return the actual column """

        # get column
        if self.columns.is_unique:
            return self._get_item_cache(key)

        # duplicate columns & possible reduce dimensionaility
        result = self._constructor(self._data.get(key))
        if result.columns.is_unique:
            result = result[key]

        return result

    def _getitem_slice(self, key):
        return self._slice(key, axis=0)

    def _getitem_array(self, key):
        # also raises Exception if object array with NA values
        if com.is_bool_indexer(key):
            # warning here just in case -- previously __setitem__ was
            # reindexing but __getitem__ was not; it seems more reasonable to
            # go with the __setitem__ behavior since that is more consistent
            # with all other indexing behavior
            if isinstance(key, Series) and not key.index.equals(self.index):
                warnings.warn("Boolean Series key will be reindexed to match "
                              "DataFrame index.", UserWarning)
            elif len(key) != len(self.index):
                raise ValueError('Item wrong length %d instead of %d.' %
                                 (len(key), len(self.index)))
            # check_bool_indexer will throw exception if Series key cannot
            # be reindexed to match DataFrame rows
            key = check_bool_indexer(self.index, key)
            indexer = key.nonzero()[0]
            return self.take(indexer, axis=0, convert=False)
        else:
            indexer = self.ix._convert_to_indexer(key, axis=1)
            return self.take(indexer, axis=1, convert=True)

    def _getitem_multilevel(self, key):
        loc = self.columns.get_loc(key)
        if isinstance(loc, (slice, Series, np.ndarray, Index)):
            new_columns = self.columns[loc]
            result_columns = maybe_droplevels(new_columns, key)
            if self._is_mixed_type:
                result = self.reindex(columns=new_columns)
                result.columns = result_columns
            else:
                new_values = self.values[:, loc]
                result = DataFrame(new_values, index=self.index,
                                   columns=result_columns).__finalize__(self)
            if len(result.columns) == 1:
                top = result.columns[0]
                if ((type(top) == str and top == '') or
                        (type(top) == tuple and top[0] == '')):
                    result = result['']
                    if isinstance(result, Series):
                        result = Series(result, index=self.index, name=key)

            result._set_is_copy(self)
            return result
        else:
            return self._get_item_cache(key)

    def _getitem_frame(self, key):
        if key.values.dtype != np.bool_:
            raise ValueError('Must pass DataFrame with boolean values only')
        return self.where(key)

    def query(self, expr, **kwargs):
        """Query the columns of a frame with a boolean expression.

        .. versionadded:: 0.13

        Parameters
        ----------
        expr : string
            The query string to evaluate.  You can refer to variables
            in the environment by prefixing them with an '@' character like
            ``@a + b``.
        kwargs : dict
            See the documentation for :func:`pandas.eval` for complete details
            on the keyword arguments accepted by :meth:`DataFrame.query`.

        Returns
        -------
        q : DataFrame

        Notes
        -----
        The result of the evaluation of this expression is first passed to
        :attr:`DataFrame.loc` and if that fails because of a
        multidimensional key (e.g., a DataFrame) then the result will be passed
        to :meth:`DataFrame.__getitem__`.

        This method uses the top-level :func:`pandas.eval` function to
        evaluate the passed query.

        The :meth:`~pandas.DataFrame.query` method uses a slightly
        modified Python syntax by default. For example, the ``&`` and ``|``
        (bitwise) operators have the precedence of their boolean cousins,
        :keyword:`and` and :keyword:`or`. This *is* syntactically valid Python,
        however the semantics are different.

        You can change the semantics of the expression by passing the keyword
        argument ``parser='python'``. This enforces the same semantics as
        evaluation in Python space. Likewise, you can pass ``engine='python'``
        to evaluate an expression using Python itself as a backend. This is not
        recommended as it is inefficient compared to using ``numexpr`` as the
        engine.

        The :attr:`DataFrame.index` and
        :attr:`DataFrame.columns` attributes of the
        :class:`~pandas.DataFrame` instance are placed in the query namespace
        by default, which allows you to treat both the index and columns of the
        frame as a column in the frame.
        The identifier ``index`` is used for the frame index; you can also
        use the name of the index to identify it in a query.

        For further details and examples see the ``query`` documentation in
        :ref:`indexing <indexing.query>`.

        See Also
        --------
        pandas.eval
        DataFrame.eval

        Examples
        --------
        >>> from numpy.random import randn
        >>> from pandas import DataFrame
        >>> df = DataFrame(randn(10, 2), columns=list('ab'))
        >>> df.query('a > b')
        >>> df[df.a > df.b]  # same result as the previous expression
        """
        kwargs['level'] = kwargs.pop('level', 0) + 1
        res = self.eval(expr, **kwargs)

        try:
            return self.loc[res]
        except ValueError:
            # when res is multi-dimensional loc raises, but this is sometimes a
            # valid query
            return self[res]

    def eval(self, expr, **kwargs):
        """Evaluate an expression in the context of the calling DataFrame
        instance.

        Parameters
        ----------
        expr : string
            The expression string to evaluate.
        kwargs : dict
            See the documentation for :func:`~pandas.eval` for complete details
            on the keyword arguments accepted by
            :meth:`~pandas.DataFrame.query`.

        Returns
        -------
        ret : ndarray, scalar, or pandas object

        See Also
        --------
        pandas.DataFrame.query
        pandas.eval

        Notes
        -----
        For more details see the API documentation for :func:`~pandas.eval`.
        For detailed examples see :ref:`enhancing performance with eval
        <enhancingperf.eval>`.

        Examples
        --------
        >>> from numpy.random import randn
        >>> from pandas import DataFrame
        >>> df = DataFrame(randn(10, 2), columns=list('ab'))
        >>> df.eval('a + b')
        >>> df.eval('c = a + b')
        """
        resolvers = kwargs.pop('resolvers', None)
        kwargs['level'] = kwargs.pop('level', 0) + 1
        if resolvers is None:
            index_resolvers = self._get_index_resolvers()
            resolvers = dict(self.iteritems()), index_resolvers
        kwargs['target'] = self
        kwargs['resolvers'] = kwargs.get('resolvers', ()) + resolvers
        return _eval(expr, **kwargs)

    def select_dtypes(self, include=None, exclude=None):
        """Return a subset of a DataFrame including/excluding columns based on
        their ``dtype``.

        Parameters
        ----------
        include, exclude : list-like
            A list of dtypes or strings to be included/excluded. You must pass
            in a non-empty sequence for at least one of these.

        Raises
        ------
        ValueError
            * If both of ``include`` and ``exclude`` are empty
            * If ``include`` and ``exclude`` have overlapping elements
            * If any kind of string dtype is passed in.
        TypeError
            * If either of ``include`` or ``exclude`` is not a sequence

        Returns
        -------
        subset : DataFrame
            The subset of the frame including the dtypes in ``include`` and
            excluding the dtypes in ``exclude``.

        Notes
        -----
        * To select all *numeric* types use the numpy dtype ``numpy.number``
        * To select strings you must use the ``object`` dtype, but note that
          this will return *all* object dtype columns
        * See the `numpy dtype hierarchy
          <http://docs.scipy.org/doc/numpy/reference/arrays.scalars.html>`__
        * To select Pandas categorical dtypes, use 'category'

        Examples
        --------
        >>> df = pd.DataFrame({'a': np.random.randn(6).astype('f4'),
        ...                    'b': [True, False] * 3,
        ...                    'c': [1.0, 2.0] * 3})
        >>> df
                a      b  c
        0  0.3962   True  1
        1  0.1459  False  2
        2  0.2623   True  1
        3  0.0764  False  2
        4 -0.9703   True  1
        5 -1.2094  False  2
        >>> df.select_dtypes(include=['float64'])
           c
        0  1
        1  2
        2  1
        3  2
        4  1
        5  2
        >>> df.select_dtypes(exclude=['floating'])
               b
        0   True
        1  False
        2   True
        3  False
        4   True
        5  False
        """
        include, exclude = include or (), exclude or ()
        if not (com.is_list_like(include) and com.is_list_like(exclude)):
            raise TypeError('include and exclude must both be non-string'
                            ' sequences')
        selection = tuple(map(frozenset, (include, exclude)))

        if not any(selection):
            raise ValueError('at least one of include or exclude must be '
                             'nonempty')

        # convert the myriad valid dtypes object to a single representation
        include, exclude = map(lambda x:
                               frozenset(map(com._get_dtype_from_object, x)),
                               selection)
        for dtypes in (include, exclude):
            com._invalidate_string_dtypes(dtypes)

        # can't both include AND exclude!
        if not include.isdisjoint(exclude):
            raise ValueError('include and exclude overlap on %s'
                             % (include & exclude))

        # empty include/exclude -> defaults to True
        # three cases (we've already raised if both are empty)
        # case 1: empty include, nonempty exclude
        # we have True, True, ... True for include, same for exclude
        # in the loop below we get the excluded
        # and when we call '&' below we get only the excluded
        # case 2: nonempty include, empty exclude
        # same as case 1, but with include
        # case 3: both nonempty
        # the "union" of the logic of case 1 and case 2:
        # we get the included and excluded, and return their logical and
        include_these = Series(not bool(include), index=self.columns)
        exclude_these = Series(not bool(exclude), index=self.columns)

        def is_dtype_instance_mapper(column, dtype):
            return column, functools.partial(issubclass, dtype.type)

        for column, f in itertools.starmap(is_dtype_instance_mapper,
                                           self.dtypes.iteritems()):
            if include:  # checks for the case of empty include or exclude
                include_these[column] = any(map(f, include))
            if exclude:
                exclude_these[column] = not any(map(f, exclude))

        dtype_indexer = include_these & exclude_these
        return self.loc[com._get_info_slice(self, dtype_indexer)]

    def _box_item_values(self, key, values):
        items = self.columns[self.columns.get_loc(key)]
        if values.ndim == 2:
            return self._constructor(values.T, columns=items, index=self.index)
        else:
            return self._box_col_values(values, items)

    def _box_col_values(self, values, items):
        """ provide boxed values for a column """
        return self._constructor_sliced.from_array(values, index=self.index,
                                                   name=items, fastpath=True)

    def __setitem__(self, key, value):

        # see if we can slice the rows
        indexer = convert_to_index_sliceable(self, key)
        if indexer is not None:
            return self._setitem_slice(indexer, value)

        if isinstance(key, (Series, np.ndarray, list, Index)):
            self._setitem_array(key, value)
        elif isinstance(key, DataFrame):
            self._setitem_frame(key, value)
        else:
            # set column
            self._set_item(key, value)

    def _setitem_slice(self, key, value):
        self._check_setitem_copy()
        self.ix._setitem_with_indexer(key, value)

    def _setitem_array(self, key, value):
        # also raises Exception if object array with NA values
        if com.is_bool_indexer(key):
            if len(key) != len(self.index):
                raise ValueError('Item wrong length %d instead of %d!' %
                                 (len(key), len(self.index)))
            key = check_bool_indexer(self.index, key)
            indexer = key.nonzero()[0]
            self._check_setitem_copy()
            self.ix._setitem_with_indexer(indexer, value)
        else:
            if isinstance(value, DataFrame):
                if len(value.columns) != len(key):
                    raise ValueError('Columns must be same length as key')
                for k1, k2 in zip(key, value.columns):
                    self[k1] = value[k2]
            else:
                indexer = self.ix._convert_to_indexer(key, axis=1)
                self._check_setitem_copy()
                self.ix._setitem_with_indexer((slice(None), indexer), value)

    def _setitem_frame(self, key, value):
        # support boolean setting with DataFrame input, e.g.
        # df[df > df2] = 0
        if key.values.dtype != np.bool_:
            raise TypeError('Must pass DataFrame with boolean values only')

        self._check_inplace_setting(value)
        self._check_setitem_copy()
        self.where(-key, value, inplace=True)

    def _ensure_valid_index(self, value):
        """
        ensure that if we don't have an index, that we can create one from the
        passed value
        """
        if not len(self.index):

            # GH5632, make sure that we are a Series convertible
            if is_list_like(value):
                try:
                    value = Series(value)
                except:
                    pass

                if not isinstance(value, Series):
                    raise ValueError('Cannot set a frame with no defined index '
                                     'and a value that cannot be converted to a '
                                     'Series')

                self._data = self._data.reindex_axis(value.index.copy(), axis=1,
                                                     fill_value=np.nan)

            # we are a scalar
            # noop
            else:

                pass

    def _set_item(self, key, value):
        """
        Add series to DataFrame in specified column.

        If series is a numpy-array (not a Series/TimeSeries), it must be the
        same length as the DataFrames index or an error will be thrown.

        Series/TimeSeries will be conformed to the DataFrames index to
        ensure homogeneity.
        """

        self._ensure_valid_index(value)
        value = self._sanitize_column(key, value)
        NDFrame._set_item(self, key, value)

        # check if we are modifying a copy
        # try to set first as we want an invalid
        # value exeption to occur first
        if len(self):
            self._check_setitem_copy()

    def insert(self, loc, column, value, allow_duplicates=False):
        """
        Insert column into DataFrame at specified location.

        If `allow_duplicates` is False, raises Exception if column
        is already contained in the DataFrame.

        Parameters
        ----------
        loc : int
            Must have 0 <= loc <= len(columns)
        column : object
        value : int, Series, or array-like
        """
        self._ensure_valid_index(value)
        value = self._sanitize_column(column, value)
        self._data.insert(
            loc, column, value, allow_duplicates=allow_duplicates)

    def _sanitize_column(self, key, value):
        # Need to make sure new columns (which go into the BlockManager as new
        # blocks) are always copied

        def reindexer(value):
            # reindex if necessary

            if value.index.equals(self.index) or not len(self.index):
                value = value.values.copy()
            else:

                # GH 4107
                try:
                    value = value.reindex(self.index).values
                except Exception as e:

                    # duplicate axis
                    if not value.index.is_unique:
                        raise e

                    # other
                    raise TypeError('incompatible index of inserted column '
                                    'with frame index')
            return value

        if isinstance(value, Series):
            value = reindexer(value)

        elif isinstance(value, DataFrame):
            # align right-hand-side columns if self.columns
            # is multi-index and self[key] is a sub-frame
            if isinstance(self.columns, MultiIndex) and key in self.columns:
                loc = self.columns.get_loc(key)
                if isinstance(loc, (slice, Series, np.ndarray, Index)):
                    cols = maybe_droplevels(self.columns[loc], key)
                    if len(cols) and not cols.equals(value.columns):
                        value = value.reindex_axis(cols, axis=1)
            # now align rows
            value = reindexer(value).T

        elif isinstance(value, Categorical):
            value = value.copy()

        elif (isinstance(value, Index) or is_sequence(value)):
            from pandas.core.series import _sanitize_index

            # turn me into an ndarray
            value = _sanitize_index(value, self.index, copy=False)
            if not isinstance(value, (np.ndarray, Index)):
                if isinstance(value, list) and len(value) > 0:
                    value = com._possibly_convert_platform(value)
                else:
                    value = com._asarray_tuplesafe(value)
            elif value.ndim == 2:
                value = value.copy().T
            else:
                value = value.copy()

            # possibly infer to datetimelike
            if is_object_dtype(value.dtype):
                value = _possibly_infer_to_datetimelike(value.ravel()).reshape(value.shape)

        else:
            # upcast the scalar
            dtype, value = _infer_dtype_from_scalar(value)
            value = np.repeat(value, len(self.index)).astype(dtype)
            value = com._possibly_cast_to_datetime(value, dtype)

        # return unconsolidatables directly
        if isinstance(value, (Categorical, SparseArray)):
            return value

        # broadcast across multiple columns if necessary
        if key in self.columns and value.ndim == 1:
            if not self.columns.is_unique or isinstance(self.columns,
                                                        MultiIndex):
                existing_piece = self[key]
                if isinstance(existing_piece, DataFrame):
                    value = np.tile(value, (len(existing_piece.columns), 1))

        return np.atleast_2d(np.asarray(value))

    @property
    def _series(self):
        result = {}
        for idx, item in enumerate(self.columns):
            result[item] = Series(self._data.iget(idx), index=self.index,
                                  name=item)
        return result

    def lookup(self, row_labels, col_labels):
        """Label-based "fancy indexing" function for DataFrame.
        Given equal-length arrays of row and column labels, return an
        array of the values corresponding to each (row, col) pair.

        Parameters
        ----------
        row_labels : sequence
            The row labels to use for lookup
        col_labels : sequence
            The column labels to use for lookup

        Notes
        -----
        Akin to::

            result = []
            for row, col in zip(row_labels, col_labels):
                result.append(df.get_value(row, col))

        Examples
        --------
        values : ndarray
            The found values

        """
        n = len(row_labels)
        if n != len(col_labels):
            raise ValueError('Row labels must have same size as column labels')

        thresh = 1000
        if not self._is_mixed_type or n > thresh:
            values = self.values
            ridx = self.index.get_indexer(row_labels)
            cidx = self.columns.get_indexer(col_labels)
            if (ridx == -1).any():
                raise KeyError('One or more row labels was not found')
            if (cidx == -1).any():
                raise KeyError('One or more column labels was not found')
            flat_index = ridx * len(self.columns) + cidx
            result = values.flat[flat_index]
        else:
            result = np.empty(n, dtype='O')
            for i, (r, c) in enumerate(zip(row_labels, col_labels)):
                result[i] = self.get_value(r, c)

        if is_object_dtype(result):
            result = lib.maybe_convert_objects(result)

        return result

    #----------------------------------------------------------------------
    # Reindexing and alignment

    def _reindex_axes(self, axes, level, limit, method, fill_value, copy):
        frame = self

        columns = axes['columns']
        if columns is not None:
            frame = frame._reindex_columns(columns, copy, level, fill_value,
                                           limit)

        index = axes['index']
        if index is not None:
            frame = frame._reindex_index(index, method, copy, level,
                                         fill_value, limit)

        return frame

    def _reindex_index(self, new_index, method, copy, level, fill_value=NA,
                       limit=None):
        new_index, indexer = self.index.reindex(new_index, method, level,
                                                limit=limit)
        return self._reindex_with_indexers({0: [new_index, indexer]},
                                           copy=copy, fill_value=fill_value,
                                           allow_dups=False)

    def _reindex_columns(self, new_columns, copy, level, fill_value=NA,
                         limit=None):
        new_columns, indexer = self.columns.reindex(new_columns, level=level,
                                                    limit=limit)
        return self._reindex_with_indexers({1: [new_columns, indexer]},
                                           copy=copy, fill_value=fill_value,
                                           allow_dups=False)

    def _reindex_multi(self, axes, copy, fill_value):
        """ we are guaranteed non-Nones in the axes! """

        new_index, row_indexer = self.index.reindex(axes['index'])
        new_columns, col_indexer = self.columns.reindex(axes['columns'])

        if row_indexer is not None and col_indexer is not None:
            indexer = row_indexer, col_indexer
            new_values = com.take_2d_multi(self.values, indexer,
                                           fill_value=fill_value)
            return self._constructor(new_values, index=new_index,
                                     columns=new_columns)
        else:
            return self._reindex_with_indexers({0: [new_index,   row_indexer],
                                                1: [new_columns, col_indexer]},
                                               copy=copy,
                                               fill_value=fill_value)

    @Appender(_shared_docs['reindex'] % _shared_doc_kwargs)
    def reindex(self, index=None, columns=None, **kwargs):
        return super(DataFrame, self).reindex(index=index, columns=columns,
                                              **kwargs)

    @Appender(_shared_docs['reindex_axis'] % _shared_doc_kwargs)
    def reindex_axis(self, labels, axis=0, method=None, level=None, copy=True,
                     limit=None, fill_value=np.nan):
        return super(DataFrame, self).reindex_axis(labels=labels, axis=axis,
                                                   method=method, level=level,
                                                   copy=copy, limit=limit,
                                                   fill_value=fill_value)

    @Appender(_shared_docs['rename'] % _shared_doc_kwargs)
    def rename(self, index=None, columns=None, **kwargs):
        return super(DataFrame, self).rename(index=index, columns=columns,
                                             **kwargs)

    def set_index(self, keys, drop=True, append=False, inplace=False,
                  verify_integrity=False):
        """
        Set the DataFrame index (row labels) using one or more existing
        columns. By default yields a new object.

        Parameters
        ----------
        keys : column label or list of column labels / arrays
        drop : boolean, default True
            Delete columns to be used as the new index
        append : boolean, default False
            Whether to append columns to existing index
        inplace : boolean, default False
            Modify the DataFrame in place (do not create a new object)
        verify_integrity : boolean, default False
            Check the new index for duplicates. Otherwise defer the check until
            necessary. Setting to False will improve the performance of this
            method

        Examples
        --------
        >>> indexed_df = df.set_index(['A', 'B'])
        >>> indexed_df2 = df.set_index(['A', [0, 1, 2, 0, 1, 2]])
        >>> indexed_df3 = df.set_index([[0, 1, 2, 0, 1, 2]])

        Returns
        -------
        dataframe : DataFrame
        """
        if not isinstance(keys, list):
            keys = [keys]

        if inplace:
            frame = self
        else:
            frame = self.copy()

        arrays = []
        names = []
        if append:
            names = [x for x in self.index.names]
            if isinstance(self.index, MultiIndex):
                for i in range(self.index.nlevels):
                    arrays.append(self.index.get_level_values(i))
            else:
                arrays.append(self.index)

        to_remove = []
        for col in keys:
            if isinstance(col, MultiIndex):
                # append all but the last column so we don't have to modify
                # the end of this loop
                for n in range(col.nlevels - 1):
                    arrays.append(col.get_level_values(n))

                level = col.get_level_values(col.nlevels - 1)
                names.extend(col.names)
            elif isinstance(col, Series):
                level = col.values
                names.append(col.name)
            elif isinstance(col, Index):
                level = col
                names.append(col.name)
            elif isinstance(col, (list, np.ndarray, Index)):
                level = col
                names.append(None)
            else:
                level = frame[col].values
                names.append(col)
                if drop:
                    to_remove.append(col)
            arrays.append(level)

        index = MultiIndex.from_arrays(arrays, names=names)

        if verify_integrity and not index.is_unique:
            duplicates = index.get_duplicates()
            raise ValueError('Index has duplicate keys: %s' % duplicates)

        for c in to_remove:
            del frame[c]

        # clear up memory usage
        index._cleanup()

        frame.index = index

        if not inplace:
            return frame

    def reset_index(self, level=None, drop=False, inplace=False, col_level=0,
                    col_fill=''):
        """
        For DataFrame with multi-level index, return new DataFrame with
        labeling information in the columns under the index names, defaulting
        to 'level_0', 'level_1', etc. if any are None. For a standard index,
        the index name will be used (if set), otherwise a default 'index' or
        'level_0' (if 'index' is already taken) will be used.

        Parameters
        ----------
        level : int, str, tuple, or list, default None
            Only remove the given levels from the index. Removes all levels by
            default
        drop : boolean, default False
            Do not try to insert index into dataframe columns. This resets
            the index to the default integer index.
        inplace : boolean, default False
            Modify the DataFrame in place (do not create a new object)
        col_level : int or str, default 0
            If the columns have multiple levels, determines which level the
            labels are inserted into. By default it is inserted into the first
            level.
        col_fill : object, default ''
            If the columns have multiple levels, determines how the other
            levels are named. If None then the index name is repeated.

        Returns
        -------
        resetted : DataFrame
        """
        if inplace:
            new_obj = self
        else:
            new_obj = self.copy()

        def _maybe_casted_values(index, labels=None):
            if isinstance(index, PeriodIndex):
                values = index.asobject.values
            elif (isinstance(index, DatetimeIndex) and
                  index.tz is not None):
                values = index.asobject
            else:
                values = index.values
                if values.dtype == np.object_:
                    values = lib.maybe_convert_objects(values)

            # if we have the labels, extract the values with a mask
            if labels is not None:
                mask = labels == -1
                values = values.take(labels)
                if mask.any():
                    values, changed = com._maybe_upcast_putmask(values,
                                                                mask, np.nan)
            return values

        new_index = np.arange(len(new_obj),dtype='int64')
        if isinstance(self.index, MultiIndex):
            if level is not None:
                if not isinstance(level, (tuple, list)):
                    level = [level]
                level = [self.index._get_level_number(lev) for lev in level]
                if len(level) < len(self.index.levels):
                    new_index = self.index.droplevel(level)

            if not drop:
                names = self.index.names
                zipped = lzip(self.index.levels, self.index.labels)

                multi_col = isinstance(self.columns, MultiIndex)
                for i, (lev, lab) in reversed(list(enumerate(zipped))):
                    col_name = names[i]
                    if col_name is None:
                        col_name = 'level_%d' % i

                    if multi_col:
                        if col_fill is None:
                            col_name = tuple([col_name] *
                                             self.columns.nlevels)
                        else:
                            name_lst = [col_fill] * self.columns.nlevels
                            lev_num = self.columns._get_level_number(col_level)
                            name_lst[lev_num] = col_name
                            col_name = tuple(name_lst)

                    # to ndarray and maybe infer different dtype
                    level_values = _maybe_casted_values(lev, lab)
                    if level is None or i in level:
                        new_obj.insert(0, col_name, level_values)

        elif not drop:
            name = self.index.name
            if name is None or name == 'index':
                name = 'index' if 'index' not in self else 'level_0'
            if isinstance(self.columns, MultiIndex):
                if col_fill is None:
                    name = tuple([name] * self.columns.nlevels)
                else:
                    name_lst = [col_fill] * self.columns.nlevels
                    lev_num = self.columns._get_level_number(col_level)
                    name_lst[lev_num] = name
                    name = tuple(name_lst)
            values = _maybe_casted_values(self.index)
            new_obj.insert(0, name, values)

        new_obj.index = new_index
        if not inplace:
            return new_obj


    #----------------------------------------------------------------------
    # Reindex-based selection methods

    def dropna(self, axis=0, how='any', thresh=None, subset=None,
               inplace=False):
        """
        Return object with labels on given axis omitted where alternately any
        or all of the data are missing

        Parameters
        ----------
        axis : {0, 1}, or tuple/list thereof
            Pass tuple or list to drop on multiple axes
        how : {'any', 'all'}
            * any : if any NA values are present, drop that label
            * all : if all values are NA, drop that label
        thresh : int, default None
            int value : require that many non-NA values
        subset : array-like
            Labels along other axis to consider, e.g. if you are dropping rows
            these would be a list of columns to include
        inplace : boolean, defalt False
            If True, do operation inplace and return None.

        Returns
        -------
        dropped : DataFrame
        """
        if isinstance(axis, (tuple, list)):
            result = self
            for ax in axis:
                result = result.dropna(how=how, thresh=thresh,
                                       subset=subset, axis=ax)
        else:
            axis = self._get_axis_number(axis)
            agg_axis = 1 - axis

            agg_obj = self
            if subset is not None:
                ax = self._get_axis(agg_axis)
                indices = ax.get_indexer_for(subset)
                check = indices == -1
                if check.any():
                    raise KeyError(list(np.compress(check,subset)))
                agg_obj = self.take(indices,axis=agg_axis)

            count = agg_obj.count(axis=agg_axis)

            if thresh is not None:
                mask = count >= thresh
            elif how == 'any':
                mask = count == len(agg_obj._get_axis(agg_axis))
            elif how == 'all':
                mask = count > 0
            else:
                if how is not None:
                    raise ValueError('invalid how option: %s' % how)
                else:
                    raise TypeError('must specify how or thresh')

            result = self.take(mask.nonzero()[0], axis=axis, convert=False)

        if inplace:
            self._update_inplace(result)
        else:
            return result

    @deprecate_kwarg(old_arg_name='cols', new_arg_name='subset')
    def drop_duplicates(self, subset=None, take_last=False, inplace=False):
        """
        Return DataFrame with duplicate rows removed, optionally only
        considering certain columns

        Parameters
        ----------
        subset : column label or sequence of labels, optional
            Only consider certain columns for identifying duplicates, by
            default use all of the columns
        take_last : boolean, default False
            Take the last observed row in a row. Defaults to the first row
        inplace : boolean, default False
            Whether to drop duplicates in place or to return a copy
        cols : kwargs only argument of subset [deprecated]

        Returns
        -------
        deduplicated : DataFrame
        """
        duplicated = self.duplicated(subset, take_last=take_last)

        if inplace:
            inds, = (-duplicated).nonzero()
            new_data = self._data.take(inds)
            self._update_inplace(new_data)
        else:
            return self[-duplicated]

    @deprecate_kwarg(old_arg_name='cols', new_arg_name='subset')
    def duplicated(self, subset=None, take_last=False):
        """
        Return boolean Series denoting duplicate rows, optionally only
        considering certain columns

        Parameters
        ----------
        subset : column label or sequence of labels, optional
            Only consider certain columns for identifying duplicates, by
            default use all of the columns
        take_last : boolean, default False
            For a set of distinct duplicate rows, flag all but the last row as
            duplicated. Default is for all but the first row to be flagged
        cols : kwargs only argument of subset [deprecated]

        Returns
        -------
        duplicated : Series
        """
        # kludge for #1833
        def _m8_to_i8(x):
            if issubclass(x.dtype.type, np.datetime64):
                return x.view(np.int64)
            return x

        # if we are only duplicating on Categoricals this can be much faster
        if subset is None:
            values = list(_m8_to_i8(self.get_values().T))
        else:
            if np.iterable(subset) and not isinstance(subset, compat.string_types):
                if isinstance(subset, tuple):
                    if subset in self.columns:
                        values = [self[subset].get_values()]
                    else:
                        values = [_m8_to_i8(self[x].get_values()) for x in subset]
                else:
                    values = [_m8_to_i8(self[x].get_values()) for x in subset]
            else:
                values = [self[subset].get_values()]

        keys = lib.fast_zip_fillna(values)
        duplicated = lib.duplicated(keys, take_last=take_last)
        return Series(duplicated, index=self.index)

    #----------------------------------------------------------------------
    # Sorting

    def sort(self, columns=None, axis=0, ascending=True,
             inplace=False, kind='quicksort', na_position='last'):
        """
        Sort DataFrame either by labels (along either axis) or by the values in
        column(s)

        Parameters
        ----------
        columns : object
            Column name(s) in frame. Accepts a column name or a list
            for a nested sort. A tuple will be interpreted as the
            levels of a multi-index.
        ascending : boolean or list, default True
            Sort ascending vs. descending. Specify list for multiple sort
            orders
        axis : {0, 1}
            Sort index/rows versus columns
        inplace : boolean, default False
            Sort the DataFrame without creating a new instance
        kind : {'quicksort', 'mergesort', 'heapsort'}, optional
            This option is only applied when sorting on a single column or label.
        na_position : {'first', 'last'} (optional, default='last')
            'first' puts NaNs at the beginning
            'last' puts NaNs at the end

        Examples
        --------
        >>> result = df.sort(['A', 'B'], ascending=[1, 0])

        Returns
        -------
        sorted : DataFrame
        """
        return self.sort_index(by=columns, axis=axis, ascending=ascending,
                               inplace=inplace, kind=kind, na_position=na_position)

    def sort_index(self, axis=0, by=None, ascending=True, inplace=False,
                   kind='quicksort', na_position='last'):
        """
        Sort DataFrame either by labels (along either axis) or by the values in
        a column

        Parameters
        ----------
        axis : {0, 1}
            Sort index/rows versus columns
        by : object
            Column name(s) in frame. Accepts a column name or a list
            for a nested sort. A tuple will be interpreted as the
            levels of a multi-index.
        ascending : boolean or list, default True
            Sort ascending vs. descending. Specify list for multiple sort
            orders
        inplace : boolean, default False
            Sort the DataFrame without creating a new instance
        na_position : {'first', 'last'} (optional, default='last')
            'first' puts NaNs at the beginning
            'last' puts NaNs at the end
        kind : {'quicksort', 'mergesort', 'heapsort'}, optional
            This option is only applied when sorting on a single column or label.

        Examples
        --------
        >>> result = df.sort_index(by=['A', 'B'], ascending=[True, False])

        Returns
        -------
        sorted : DataFrame
        """

        from pandas.core.groupby import _lexsort_indexer, _nargsort
        axis = self._get_axis_number(axis)
        if axis not in [0, 1]:  # pragma: no cover
            raise AssertionError('Axis must be 0 or 1, got %s' % str(axis))

        labels = self._get_axis(axis)

        if by is not None:
            if axis != 0:
                raise ValueError('When sorting by column, axis must be 0 '
                                 '(rows)')
            if not isinstance(by, list):
                by = [by]
            if com.is_sequence(ascending) and len(by) != len(ascending):
                raise ValueError('Length of ascending (%d) != length of by'
                                 ' (%d)' % (len(ascending), len(by)))
            if len(by) > 1:
                def trans(v):
                    if com.needs_i8_conversion(v):
                        return v.view('i8')
                    return v
                keys = []
                for x in by:
                    k = self[x].values
                    if k.ndim == 2:
                        raise ValueError('Cannot sort by duplicate column %s' % str(x))
                    keys.append(trans(k))
                indexer = _lexsort_indexer(keys, orders=ascending,
                                           na_position=na_position)
                indexer = com._ensure_platform_int(indexer)
            else:
                by = by[0]
                k = self[by].values
                if k.ndim == 2:

                    # try to be helpful
                    if isinstance(self.columns, MultiIndex):
                        raise ValueError('Cannot sort by column %s in a multi-index'
                                         '  you need to explicity provide all the levels'
                                         % str(by))

                    raise ValueError('Cannot sort by duplicate column %s'
                                     % str(by))
                if isinstance(ascending, (tuple, list)):
                    ascending = ascending[0]

                indexer = _nargsort(k, kind=kind, ascending=ascending,
                                    na_position=na_position)

        elif isinstance(labels, MultiIndex):

            # make sure that the axis is lexsorted to start
            # if not we need to reconstruct to get the correct indexer
            if not labels.is_lexsorted():
                labels = MultiIndex.from_tuples(labels.values)

            indexer = _lexsort_indexer(labels.labels, orders=ascending,
                                       na_position=na_position)
            indexer = com._ensure_platform_int(indexer)
        else:
            indexer = _nargsort(labels, kind=kind, ascending=ascending,
                                na_position=na_position)

        bm_axis = self._get_block_manager_axis(axis)
        new_data = self._data.take(indexer, axis=bm_axis,
                                   convert=False, verify=False)

        if inplace:
            return self._update_inplace(new_data)
        else:
            return self._constructor(new_data).__finalize__(self)

    def sortlevel(self, level=0, axis=0, ascending=True,
                  inplace=False, sort_remaining=True):
        """
        Sort multilevel index by chosen axis and primary level. Data will be
        lexicographically sorted by the chosen level followed by the other
        levels (in order)

        Parameters
        ----------
        level : int
        axis : {0, 1}
        ascending : boolean, default True
        inplace : boolean, default False
            Sort the DataFrame without creating a new instance
        sort_remaining : boolean, default True
            Sort by the other levels too.

        Returns
        -------
        sorted : DataFrame
        """
        axis = self._get_axis_number(axis)
        the_axis = self._get_axis(axis)
        if not isinstance(the_axis, MultiIndex):
            raise TypeError('can only sort by level with a hierarchical index')

        new_axis, indexer = the_axis.sortlevel(level, ascending=ascending,
                                               sort_remaining=sort_remaining)

        if self._is_mixed_type and not inplace:
            ax = 'index' if axis == 0 else 'columns'

            if new_axis.is_unique:
                return self.reindex(**{ax: new_axis})
            else:
                return self.take(indexer, axis=axis, convert=False)

        bm_axis = self._get_block_manager_axis(axis)
        new_data = self._data.take(indexer, axis=bm_axis,
                                   convert=False, verify=False)
        if inplace:
            return self._update_inplace(new_data)
        else:
            return self._constructor(new_data).__finalize__(self)

    def swaplevel(self, i, j, axis=0):
        """
        Swap levels i and j in a MultiIndex on a particular axis

        Parameters
        ----------
        i, j : int, string (can be mixed)
            Level of index to be swapped. Can pass level name as string.

        Returns
        -------
        swapped : type of caller (new object)
        """
        result = self.copy()

        axis = self._get_axis_number(axis)
        if axis == 0:
            result.index = result.index.swaplevel(i, j)
        else:
            result.columns = result.columns.swaplevel(i, j)
        return result

    def reorder_levels(self, order, axis=0):
        """
        Rearrange index levels using input order.
        May not drop or duplicate levels

        Parameters
        ----------
        order : list of int or list of str
            List representing new level order. Reference level by number
            (position) or by key (label).
        axis : int
            Where to reorder levels.

        Returns
        -------
        type of caller (new object)
        """
        axis = self._get_axis_number(axis)
        if not isinstance(self._get_axis(axis),
                          MultiIndex):  # pragma: no cover
            raise TypeError('Can only reorder levels on a hierarchical axis.')

        result = self.copy()

        if axis == 0:
            result.index = result.index.reorder_levels(order)
        else:
            result.columns = result.columns.reorder_levels(order)
        return result

    #----------------------------------------------------------------------
    # Arithmetic / combination related

    def _combine_frame(self, other, func, fill_value=None, level=None):
        this, other = self.align(other, join='outer', level=level, copy=False)
        new_index, new_columns = this.index, this.columns

        def _arith_op(left, right):
            if fill_value is not None:
                left_mask = isnull(left)
                right_mask = isnull(right)
                left = left.copy()
                right = right.copy()

                # one but not both
                mask = left_mask ^ right_mask
                left[left_mask & mask] = fill_value
                right[right_mask & mask] = fill_value

            return func(left, right)

        if this._is_mixed_type or other._is_mixed_type:

            # unique
            if this.columns.is_unique:

                def f(col):
                    r = _arith_op(this[col].values, other[col].values)
                    return self._constructor_sliced(r, index=new_index,
                                                    dtype=r.dtype)

                result = dict([(col, f(col)) for col in this])

            # non-unique
            else:

                def f(i):
                    r = _arith_op(this.iloc[:, i].values,
                                  other.iloc[:, i].values)
                    return self._constructor_sliced(r, index=new_index,
                                                    dtype=r.dtype)

                result = dict([
                    (i, f(i)) for i, col in enumerate(this.columns)
                ])
                result = self._constructor(result, index=new_index, copy=False)
                result.columns = new_columns
                return result

        else:
            result = _arith_op(this.values, other.values)

        return self._constructor(result, index=new_index,
                                 columns=new_columns, copy=False)

    def _combine_series(self, other, func, fill_value=None, axis=None,
                        level=None):
        if axis is not None:
            axis = self._get_axis_name(axis)
            if axis == 'index':
                return self._combine_match_index(other, func, level=level, fill_value=fill_value)
            else:
                return self._combine_match_columns(other, func, level=level, fill_value=fill_value)
        return self._combine_series_infer(other, func, level=level, fill_value=fill_value)

    def _combine_series_infer(self, other, func, level=None, fill_value=None):
        if len(other) == 0:
            return self * NA

        if len(self) == 0:
            # Ambiguous case, use _series so works with DataFrame
            return self._constructor(data=self._series, index=self.index,
                                     columns=self.columns)

        # teeny hack because one does DataFrame + TimeSeries all the time
        if self.index.is_all_dates and other.index.is_all_dates:
            warnings.warn(("TimeSeries broadcasting along DataFrame index "
                           "by default is deprecated. Please use "
                           "DataFrame.<op> to explicitly broadcast arithmetic "
                           "operations along the index"),
                          FutureWarning)
            return self._combine_match_index(other, func, level=level, fill_value=fill_value)
        else:
            return self._combine_match_columns(other, func, level=level, fill_value=fill_value)

    def _combine_match_index(self, other, func, level=None, fill_value=None):
        left, right = self.align(other, join='outer', axis=0, level=level, copy=False)
        if fill_value is not None:
            raise NotImplementedError("fill_value %r not supported." %
                                      fill_value)
        return self._constructor(func(left.values.T, right.values).T,
                                 index=left.index,
                                 columns=self.columns, copy=False)

    def _combine_match_columns(self, other, func, level=None, fill_value=None):
        left, right = self.align(other, join='outer', axis=1, level=level, copy=False)
        if fill_value is not None:
            raise NotImplementedError("fill_value %r not supported" %
                                      fill_value)

        new_data = left._data.eval(
            func=func, other=right, axes=[left.columns, self.index])
        return self._constructor(new_data)

    def _combine_const(self, other, func, raise_on_error=True):
        if self.empty:
            return self

        new_data = self._data.eval(func=func, other=other, raise_on_error=raise_on_error)
        return self._constructor(new_data)

    def _compare_frame_evaluate(self, other, func, str_rep):

        # unique
        if self.columns.is_unique:
            def _compare(a, b):
                return dict([(col, func(a[col], b[col])) for col in a.columns])
            new_data = expressions.evaluate(_compare, str_rep, self, other)
            return self._constructor(data=new_data, index=self.index,
                                     columns=self.columns, copy=False)
        # non-unique
        else:
            def _compare(a, b):
                return dict([(i, func(a.iloc[:, i], b.iloc[:, i]))
                             for i, col in enumerate(a.columns)])
            new_data = expressions.evaluate(_compare, str_rep, self, other)
            result = self._constructor(data=new_data, index=self.index,
                                       copy=False)
            result.columns = self.columns
            return result

    def _compare_frame(self, other, func, str_rep):
        if not self._indexed_same(other):
            raise ValueError('Can only compare identically-labeled '
                             'DataFrame objects')
        return self._compare_frame_evaluate(other, func, str_rep)

    def _flex_compare_frame(self, other, func, str_rep, level):
        if not self._indexed_same(other):
            self, other = self.align(other, 'outer', level=level, copy=False)
        return self._compare_frame_evaluate(other, func, str_rep)

    def combine(self, other, func, fill_value=None, overwrite=True):
        """
        Add two DataFrame objects and do not propagate NaN values, so if for a
        (column, time) one frame is missing a value, it will default to the
        other frame's value (which might be NaN as well)

        Parameters
        ----------
        other : DataFrame
        func : function
        fill_value : scalar value
        overwrite : boolean, default True
            If True then overwrite values for common keys in the calling frame

        Returns
        -------
        result : DataFrame
        """

        other_idxlen = len(other.index)  # save for compare

        this, other = self.align(other, copy=False)
        new_index = this.index

        if other.empty and len(new_index) == len(self.index):
            return self.copy()

        if self.empty and len(other) == other_idxlen:
            return other.copy()

        # sorts if possible
        new_columns = this.columns.union(other.columns)
        do_fill = fill_value is not None

        result = {}
        for col in new_columns:
            series = this[col]
            otherSeries = other[col]

            this_dtype = series.dtype
            other_dtype = otherSeries.dtype

            this_mask = isnull(series)
            other_mask = isnull(otherSeries)

            # don't overwrite columns unecessarily
            # DO propogate if this column is not in the intersection
            if not overwrite and other_mask.all():
                result[col] = this[col].copy()
                continue

            if do_fill:
                series = series.copy()
                otherSeries = otherSeries.copy()
                series[this_mask] = fill_value
                otherSeries[other_mask] = fill_value

            # if we have different dtypes, possibily promote
            new_dtype = this_dtype
            if this_dtype != other_dtype:
                new_dtype = com._lcd_dtypes(this_dtype, other_dtype)
                series = series.astype(new_dtype)
                otherSeries = otherSeries.astype(new_dtype)

            # see if we need to be represented as i8 (datetimelike)
            # try to keep us at this dtype
            needs_i8_conversion = com.needs_i8_conversion(new_dtype)
            if needs_i8_conversion:
                this_dtype = new_dtype
                arr = func(series, otherSeries, True)
            else:
                arr = func(series, otherSeries)

            if do_fill:
                arr = com.ensure_float(arr)
                arr[this_mask & other_mask] = NA

            # try to downcast back to the original dtype
            if needs_i8_conversion:
                arr = com._possibly_cast_to_datetime(arr, this_dtype)
            else:
                arr = com._possibly_downcast_to_dtype(arr, this_dtype)

            result[col] = arr

        # convert_objects just in case
        return self._constructor(result,
                                 index=new_index,
                                 columns=new_columns).convert_objects(
            convert_dates=True,
            copy=False)

    def combine_first(self, other):
        """
        Combine two DataFrame objects and default to non-null values in frame
        calling the method. Result index columns will be the union of the
        respective indexes and columns

        Parameters
        ----------
        other : DataFrame

        Examples
        --------
        a's values prioritized, use values from b to fill holes:

        >>> a.combine_first(b)


        Returns
        -------
        combined : DataFrame
        """
        def combiner(x, y, needs_i8_conversion=False):
            x_values = x.values if hasattr(x, 'values') else x
            y_values = y.values if hasattr(y, 'values') else y
            if needs_i8_conversion:
                mask = isnull(x)
                x_values = x_values.view('i8')
                y_values = y_values.view('i8')
            else:
                mask = isnull(x_values)

            return expressions.where(mask, y_values, x_values,
                                     raise_on_error=True)

        return self.combine(other, combiner, overwrite=False)

    def update(self, other, join='left', overwrite=True, filter_func=None,
               raise_conflict=False):
        """
        Modify DataFrame in place using non-NA values from passed
        DataFrame. Aligns on indices

        Parameters
        ----------
        other : DataFrame, or object coercible into a DataFrame
        join : {'left'}, default 'left'
        overwrite : boolean, default True
            If True then overwrite values for common keys in the calling frame
        filter_func : callable(1d-array) -> 1d-array<boolean>, default None
            Can choose to replace values other than NA. Return True for values
            that should be updated
        raise_conflict : boolean
            If True, will raise an error if the DataFrame and other both
            contain data in the same place.
        """
        # TODO: Support other joins
        if join != 'left':  # pragma: no cover
            raise NotImplementedError("Only left join is supported")

        if not isinstance(other, DataFrame):
            other = DataFrame(other)

        other = other.reindex_like(self)

        for col in self.columns:
            this = self[col].values
            that = other[col].values
            if filter_func is not None:
                mask = ~filter_func(this) | isnull(that)
            else:
                if raise_conflict:
                    mask_this = notnull(that)
                    mask_that = notnull(this)
                    if any(mask_this & mask_that):
                        raise ValueError("Data overlaps.")

                if overwrite:
                    mask = isnull(that)

                    # don't overwrite columns unecessarily
                    if mask.all():
                        continue
                else:
                    mask = notnull(this)

            self[col] = expressions.where(
                mask, this, that, raise_on_error=True)

    #----------------------------------------------------------------------
    # Misc methods

    def first_valid_index(self):
        """
        Return label for first non-NA/null value
        """
        return self.index[self.count(1) > 0][0]

    def last_valid_index(self):
        """
        Return label for last non-NA/null value
        """
        return self.index[self.count(1) > 0][-1]

    #----------------------------------------------------------------------
    # Data reshaping

    def pivot(self, index=None, columns=None, values=None):
        """
        Reshape data (produce a "pivot" table) based on column values. Uses
        unique values from index / columns to form axes and return either
        DataFrame or Panel, depending on whether you request a single value
        column (DataFrame) or all columns (Panel)

        Parameters
        ----------
        index : string or object
            Column name to use to make new frame's index
        columns : string or object
            Column name to use to make new frame's columns
        values : string or object, optional
            Column name to use for populating new frame's values

        Notes
        -----
        For finer-tuned control, see hierarchical indexing documentation along
        with the related stack/unstack methods

        Examples
        --------
        >>> df
            foo   bar  baz
        0   one   A    1.
        1   one   B    2.
        2   one   C    3.
        3   two   A    4.
        4   two   B    5.
        5   two   C    6.

        >>> df.pivot('foo', 'bar', 'baz')
             A   B   C
        one  1   2   3
        two  4   5   6

        >>> df.pivot('foo', 'bar')['baz']
             A   B   C
        one  1   2   3
        two  4   5   6

        Returns
        -------
        pivoted : DataFrame
            If no values column specified, will have hierarchically indexed
            columns
        """
        from pandas.core.reshape import pivot
        return pivot(self, index=index, columns=columns, values=values)

    def stack(self, level=-1, dropna=True):
        """
        Pivot a level of the (possibly hierarchical) column labels, returning a
        DataFrame (or Series in the case of an object with a single level of
        column labels) having a hierarchical index with a new inner-most level
        of row labels.
        The level involved will automatically get sorted.

        Parameters
        ----------
        level : int, string, or list of these, default last level
            Level(s) to stack, can pass level name
        dropna : boolean, default True
            Whether to drop rows in the resulting Frame/Series with no valid
            values

        Examples
        ----------
        >>> s
             a   b
        one  1.  2.
        two  3.  4.

        >>> s.stack()
        one a    1
            b    2
        two a    3
            b    4

        Returns
        -------
        stacked : DataFrame or Series
        """
        from pandas.core.reshape import stack, stack_multiple

        if isinstance(level, (tuple, list)):
            return stack_multiple(self, level, dropna=dropna)
        else:
            return stack(self, level, dropna=dropna)

    def unstack(self, level=-1):
        """
        Pivot a level of the (necessarily hierarchical) index labels, returning
        a DataFrame having a new level of column labels whose inner-most level
        consists of the pivoted index labels. If the index is not a MultiIndex,
        the output will be a Series (the analogue of stack when the columns are
        not a MultiIndex).
        The level involved will automatically get sorted.

        Parameters
        ----------
        level : int, string, or list of these, default -1 (last level)
            Level(s) of index to unstack, can pass level name

        See also
        --------
        DataFrame.pivot : Pivot a table based on column values.
        DataFrame.stack : Pivot a level of the column labels (inverse operation
            from `unstack`).

        Examples
        --------
        >>> index = pd.MultiIndex.from_tuples([('one', 'a'), ('one', 'b'),
        ...                                    ('two', 'a'), ('two', 'b')])
        >>> s = pd.Series(np.arange(1.0, 5.0), index=index)
        >>> s
        one  a   1
             b   2
        two  a   3
             b   4
        dtype: float64

        >>> s.unstack(level=-1)
             a   b
        one  1  2
        two  3  4

        >>> s.unstack(level=0)
           one  two
        a  1   3
        b  2   4

        >>> df = s.unstack(level=0)
        >>> df.unstack()
        one  a  1.
             b  3.
        two  a  2.
             b  4.

        Returns
        -------
        unstacked : DataFrame or Series
        """
        from pandas.core.reshape import unstack
        return unstack(self, level)

    #----------------------------------------------------------------------
    # Time series-related

    def diff(self, periods=1):
        """
        1st discrete difference of object

        Parameters
        ----------
        periods : int, default 1
            Periods to shift for forming difference

        Returns
        -------
        diffed : DataFrame
        """
        new_data = self._data.diff(n=periods)
        return self._constructor(new_data)

    #----------------------------------------------------------------------
    # Function application

    def apply(self, func, axis=0, broadcast=False, raw=False, reduce=None,
              args=(), **kwds):
        """
        Applies function along input axis of DataFrame.

        Objects passed to functions are Series objects having index
        either the DataFrame's index (axis=0) or the columns (axis=1).
        Return type depends on whether passed function aggregates, or the
        reduce argument if the DataFrame is empty.

        Parameters
        ----------
        func : function
            Function to apply to each column/row
        axis : {0, 1}
            * 0 : apply function to each column
            * 1 : apply function to each row
        broadcast : boolean, default False
            For aggregation functions, return object of same size with values
            propagated
        reduce : boolean or None, default None
            Try to apply reduction procedures. If the DataFrame is empty,
            apply will use reduce to determine whether the result should be a
            Series or a DataFrame. If reduce is None (the default), apply's
            return value will be guessed by calling func an empty Series (note:
            while guessing, exceptions raised by func will be ignored). If
            reduce is True a Series will always be returned, and if False a
            DataFrame will always be returned.
        raw : boolean, default False
            If False, convert each row or column into a Series. If raw=True the
            passed function will receive ndarray objects instead. If you are
            just applying a NumPy reduction function this will achieve much
            better performance
        args : tuple
            Positional arguments to pass to function in addition to the
            array/series
        Additional keyword arguments will be passed as keywords to the function

        Notes
        -----
        In the current implementation apply calls func twice on the
        first column/row to decide whether it can take a fast or slow
        code path. This can lead to unexpected behavior if func has
        side-effects, as they will take effect twice for the first
        column/row.

        Examples
        --------
        >>> df.apply(numpy.sqrt) # returns DataFrame
        >>> df.apply(numpy.sum, axis=0) # equiv to df.sum(0)
        >>> df.apply(numpy.sum, axis=1) # equiv to df.sum(1)

        See also
        --------
        DataFrame.applymap: For elementwise operations

        Returns
        -------
        applied : Series or DataFrame
        """
        axis = self._get_axis_number(axis)
        if kwds or args and not isinstance(func, np.ufunc):
            f = lambda x: func(x, *args, **kwds)
        else:
            f = func

        if len(self.columns) == 0 and len(self.index) == 0:
            return self._apply_empty_result(func, axis, reduce, *args, **kwds)

        if isinstance(f, np.ufunc):
            results = f(self.values)
            return self._constructor(data=results, index=self.index,
                                     columns=self.columns, copy=False)
        else:
            if not broadcast:
                if not all(self.shape):
                    return self._apply_empty_result(func, axis, reduce, *args,
                                                    **kwds)

                if raw and not self._is_mixed_type:
                    return self._apply_raw(f, axis)
                else:
                    if reduce is None:
                        reduce = True
                    return self._apply_standard(f, axis, reduce=reduce)
            else:
                return self._apply_broadcast(f, axis)

    def _apply_empty_result(self, func, axis, reduce, *args, **kwds):
        if reduce is None:
            reduce = False
            try:
                reduce = not isinstance(func(_EMPTY_SERIES, *args, **kwds),
                                        Series)
            except Exception:
                pass

        if reduce:
            return Series(NA, index=self._get_agg_axis(axis))
        else:
            return self.copy()

    def _apply_raw(self, func, axis):
        try:
            result = lib.reduce(self.values, func, axis=axis)
        except Exception:
            result = np.apply_along_axis(func, axis, self.values)

        # TODO: mixed type case
        if result.ndim == 2:
            return DataFrame(result, index=self.index,
                             columns=self.columns)
        else:
            return Series(result, index=self._get_agg_axis(axis))

    def _apply_standard(self, func, axis, ignore_failures=False, reduce=True):

        # skip if we are mixed datelike and trying reduce across axes
        # GH6125
        if reduce and axis==1 and self._is_mixed_type and self._is_datelike_mixed_type:
            reduce=False

        # try to reduce first (by default)
        # this only matters if the reduction in values is of different dtype
        # e.g. if we want to apply to a SparseFrame, then can't directly reduce
        if reduce:

            try:

                # the is the fast-path
                values = self.values
                dummy = Series(NA, index=self._get_axis(axis),
                               dtype=values.dtype)

                labels = self._get_agg_axis(axis)
                result = lib.reduce(values, func, axis=axis, dummy=dummy,
                                    labels=labels)
                return Series(result, index=labels)
            except Exception:
                pass

        dtype = object if self._is_mixed_type else None
        if axis == 0:
            series_gen = (self.icol(i) for i in range(len(self.columns)))
            res_index = self.columns
            res_columns = self.index
        elif axis == 1:
            res_index = self.index
            res_columns = self.columns
            values = self.values
            series_gen = (Series.from_array(arr, index=res_columns, name=name, dtype=dtype)
                          for i, (arr, name) in
                          enumerate(zip(values, res_index)))
        else:  # pragma : no cover
            raise AssertionError('Axis must be 0 or 1, got %s' % str(axis))

        i = None
        keys = []
        results = {}
        if ignore_failures:
            successes = []
            for i, v in enumerate(series_gen):
                try:
                    results[i] = func(v)
                    keys.append(v.name)
                    successes.append(i)
                except Exception:
                    pass
            # so will work with MultiIndex
            if len(successes) < len(res_index):
                res_index = res_index.take(successes)
        else:
            try:
                for i, v in enumerate(series_gen):
                    results[i] = func(v)
                    keys.append(v.name)
            except Exception as e:
                if hasattr(e, 'args'):
                    # make sure i is defined
                    if i is not None:
                        k = res_index[i]
                        e.args = e.args + ('occurred at index %s' %
                                           com.pprint_thing(k),)
                raise

        if len(results) > 0 and is_sequence(results[0]):
            if not isinstance(results[0], Series):
                index = res_columns
            else:
                index = None

            result = self._constructor(data=results, index=index)
            result.columns = res_index

            if axis == 1:
                result = result.T
            result = result.convert_objects(copy=False)

        else:

            result = Series(results)
            result.index = res_index

        return result

    def _apply_broadcast(self, func, axis):
        if axis == 0:
            target = self
        elif axis == 1:
            target = self.T
        else:  # pragma: no cover
            raise AssertionError('Axis must be 0 or 1, got %s' % axis)

        result_values = np.empty_like(target.values)
        columns = target.columns
        for i, col in enumerate(columns):
            result_values[:, i] = func(target[col])

        result = self._constructor(result_values, index=target.index,
                                   columns=target.columns)

        if axis == 1:
            result = result.T

        return result

    def applymap(self, func):
        """
        Apply a function to a DataFrame that is intended to operate
        elementwise, i.e. like doing map(func, series) for each series in the
        DataFrame

        Parameters
        ----------
        func : function
            Python function, returns a single value from a single value

        Returns
        -------
        applied : DataFrame

        See also
        --------
        DataFrame.apply : For operations on rows/columns

        """

        # if we have a dtype == 'M8[ns]', provide boxed values
        def infer(x):
            if com.needs_i8_conversion(x):
                f = com.i8_boxer(x)
                x = lib.map_infer(_values_from_object(x), f)
            return lib.map_infer(_values_from_object(x), func)
        return self.apply(infer)

    #----------------------------------------------------------------------
    # Merging / joining methods

    def append(self, other, ignore_index=False, verify_integrity=False):
        """
        Append rows of `other` to the end of this frame, returning a new
        object. Columns not in this frame are added as new columns.

        Parameters
        ----------
        other : DataFrame or Series/dict-like object, or list of these
            The data to append.
        ignore_index : boolean, default False
            If True, do not use the index labels.
        verify_integrity : boolean, default False
            If True, raise ValueError on creating index with duplicates.

        Returns
        -------
        appended : DataFrame

        Notes
        -----
        If a list of dict/series is passed and the keys are all contained in the
        DataFrame's index, the order of the columns in the resulting DataFrame
        will be unchanged.

        See also
        --------
        pandas.concat : General function to concatenate DataFrame, Series
            or Panel objects

        Examples
        --------

        >>> df = pd.DataFrame([[1, 2], [3, 4]], columns=list('AB'))
        >>> df
           A  B
        0  1  2
        1  3  4
        >>> df2 = pd.DataFrame([[5, 6], [7, 8]], columns=list('AB'))
        >>> df.append(df2)
           A  B
        0  1  2
        1  3  4
        0  5  6
        1  7  8

        With `ignore_index` set to True:

        >>> df.append(df2, ignore_index=True)
           A  B
        0  1  2
        1  3  4
        2  5  6
        3  7  8

        """
        if isinstance(other, (Series, dict)):
            if isinstance(other, dict):
                other = Series(other)
            if other.name is None and not ignore_index:
                raise TypeError('Can only append a Series if ignore_index=True'
                                ' or if the Series has a name')

            index = None if other.name is None else [other.name]
            combined_columns = self.columns.tolist() + self.columns.union(other.index).difference(self.columns).tolist()
            other = other.reindex(combined_columns, copy=False)
            other = DataFrame(other.values.reshape((1, len(other))),
                              index=index, columns=combined_columns).convert_objects()
            if not self.columns.equals(combined_columns):
                self = self.reindex(columns=combined_columns)
        elif isinstance(other, list) and not isinstance(other[0], DataFrame):
            other = DataFrame(other)
            if (self.columns.get_indexer(other.columns) >= 0).all():
                other = other.ix[:, self.columns]

        from pandas.tools.merge import concat
        if isinstance(other, (list, tuple)):
            to_concat = [self] + other
        else:
            to_concat = [self, other]
        return concat(to_concat, ignore_index=ignore_index,
                      verify_integrity=verify_integrity)

    def join(self, other, on=None, how='left', lsuffix='', rsuffix='',
             sort=False):
        """
        Join columns with other DataFrame either on index or on a key
        column. Efficiently Join multiple DataFrame objects by index at once by
        passing a list.

        Parameters
        ----------
        other : DataFrame, Series with name field set, or list of DataFrame
            Index should be similar to one of the columns in this one. If a
            Series is passed, its name attribute must be set, and that will be
            used as the column name in the resulting joined DataFrame
        on : column name, tuple/list of column names, or array-like
            Column(s) to use for joining, otherwise join on index. If multiples
            columns given, the passed DataFrame must have a MultiIndex. Can
            pass an array as the join key if not already contained in the
            calling DataFrame. Like an Excel VLOOKUP operation
        how : {'left', 'right', 'outer', 'inner'}
            How to handle indexes of the two objects. Default: 'left'
            for joining on index, None otherwise

            * left: use calling frame's index
            * right: use input frame's index
            * outer: form union of indexes
            * inner: use intersection of indexes
        lsuffix : string
            Suffix to use from left frame's overlapping columns
        rsuffix : string
            Suffix to use from right frame's overlapping columns
        sort : boolean, default False
            Order result DataFrame lexicographically by the join key. If False,
            preserves the index order of the calling (left) DataFrame

        Notes
        -----
        on, lsuffix, and rsuffix options are not supported when passing a list
        of DataFrame objects

        Returns
        -------
        joined : DataFrame
        """
        # For SparseDataFrame's benefit
        return self._join_compat(other, on=on, how=how, lsuffix=lsuffix,
                                 rsuffix=rsuffix, sort=sort)

    def _join_compat(self, other, on=None, how='left', lsuffix='', rsuffix='',
                     sort=False):
        from pandas.tools.merge import merge, concat

        if isinstance(other, Series):
            if other.name is None:
                raise ValueError('Other Series must have a name')
            other = DataFrame({other.name: other})

        if isinstance(other, DataFrame):
            return merge(self, other, left_on=on, how=how,
                         left_index=on is None, right_index=True,
                         suffixes=(lsuffix, rsuffix), sort=sort)
        else:
            if on is not None:
                raise ValueError('Joining multiple DataFrames only supported'
                                 ' for joining on index')

            # join indexes only using concat
            if how == 'left':
                how = 'outer'
                join_axes = [self.index]
            else:
                join_axes = None

            frames = [self] + list(other)

            can_concat = all(df.index.is_unique for df in frames)

            if can_concat:
                return concat(frames, axis=1, join=how, join_axes=join_axes,
                              verify_integrity=True)

            joined = frames[0]

            for frame in frames[1:]:
                joined = merge(joined, frame, how=how,
                               left_index=True, right_index=True)

            return joined

    @Substitution('')
    @Appender(_merge_doc, indents=2)
    def merge(self, right, how='inner', on=None, left_on=None, right_on=None,
              left_index=False, right_index=False, sort=False,
              suffixes=('_x', '_y'), copy=True):
        from pandas.tools.merge import merge
        return merge(self, right, how=how, on=on,
                     left_on=left_on, right_on=right_on,
                     left_index=left_index, right_index=right_index, sort=sort,
                     suffixes=suffixes, copy=copy)

    #----------------------------------------------------------------------
    # Statistical methods, etc.

    def corr(self, method='pearson', min_periods=1):
        """
        Compute pairwise correlation of columns, excluding NA/null values

        Parameters
        ----------
        method : {'pearson', 'kendall', 'spearman'}
            * pearson : standard correlation coefficient
            * kendall : Kendall Tau correlation coefficient
            * spearman : Spearman rank correlation
        min_periods : int, optional
            Minimum number of observations required per pair of columns
            to have a valid result. Currently only available for pearson
            and spearman correlation

        Returns
        -------
        y : DataFrame
        """
        numeric_df = self._get_numeric_data()
        cols = numeric_df.columns
        mat = numeric_df.values

        if method == 'pearson':
            correl = _algos.nancorr(com._ensure_float64(mat),
                                    minp=min_periods)
        elif method == 'spearman':
            correl = _algos.nancorr_spearman(com._ensure_float64(mat),
                                             minp=min_periods)
        else:
            if min_periods is None:
                min_periods = 1
            mat = mat.T
            corrf = nanops.get_corr_func(method)
            K = len(cols)
            correl = np.empty((K, K), dtype=float)
            mask = np.isfinite(mat)
            for i, ac in enumerate(mat):
                for j, bc in enumerate(mat):
                    valid = mask[i] & mask[j]
                    if valid.sum() < min_periods:
                        c = NA
                    elif not valid.all():
                        c = corrf(ac[valid], bc[valid])
                    else:
                        c = corrf(ac, bc)
                    correl[i, j] = c
                    correl[j, i] = c

        return self._constructor(correl, index=cols, columns=cols)

    def cov(self, min_periods=None):
        """
        Compute pairwise covariance of columns, excluding NA/null values

        Parameters
        ----------
        min_periods : int, optional
            Minimum number of observations required per pair of columns
            to have a valid result.

        Returns
        -------
        y : DataFrame

        Notes
        -----
        `y` contains the covariance matrix of the DataFrame's time series.
        The covariance is normalized by N-1 (unbiased estimator).
        """
        numeric_df = self._get_numeric_data()
        cols = numeric_df.columns
        mat = numeric_df.values

        if notnull(mat).all():
            if min_periods is not None and min_periods > len(mat):
                baseCov = np.empty((mat.shape[1], mat.shape[1]))
                baseCov.fill(np.nan)
            else:
                baseCov = np.cov(mat.T)
            baseCov = baseCov.reshape((len(cols), len(cols)))
        else:
            baseCov = _algos.nancorr(com._ensure_float64(mat), cov=True,
                                     minp=min_periods)

        return self._constructor(baseCov, index=cols, columns=cols)

    def corrwith(self, other, axis=0, drop=False):
        """
        Compute pairwise correlation between rows or columns of two DataFrame
        objects.

        Parameters
        ----------
        other : DataFrame
        axis : {0, 1}
            0 to compute column-wise, 1 for row-wise
        drop : boolean, default False
            Drop missing indices from result, default returns union of all

        Returns
        -------
        correls : Series
        """
        axis = self._get_axis_number(axis)
        if isinstance(other, Series):
            return self.apply(other.corr, axis=axis)

        this = self._get_numeric_data()
        other = other._get_numeric_data()

        left, right = this.align(other, join='inner', copy=False)

        # mask missing values
        left = left + right * 0
        right = right + left * 0

        if axis == 1:
            left = left.T
            right = right.T

        # demeaned data
        ldem = left - left.mean()
        rdem = right - right.mean()

        num = (ldem * rdem).sum()
        dom = (left.count() - 1) * left.std() * right.std()

        correl = num / dom

        if not drop:
            raxis = 1 if axis == 0 else 0
            result_index = this._get_axis(raxis).union(other._get_axis(raxis))
            correl = correl.reindex(result_index)

        return correl

    #----------------------------------------------------------------------
    # ndarray-like stats methods

    def count(self, axis=0, level=None, numeric_only=False):
        """
        Return Series with number of non-NA/null observations over requested
        axis. Works with non-floating point data as well (detects NaN and None)

        Parameters
        ----------
        axis : {0, 1}
            0 for row-wise, 1 for column-wise
        level : int or level name, default None
            If the axis is a MultiIndex (hierarchical), count along a
            particular level, collapsing into a DataFrame
        numeric_only : boolean, default False
            Include only float, int, boolean data

        Returns
        -------
        count : Series (or DataFrame if level specified)
        """
        axis = self._get_axis_number(axis)
        if level is not None:
            return self._count_level(level, axis=axis,
                                     numeric_only=numeric_only)

        if numeric_only:
            frame = self._get_numeric_data()
        else:
            frame = self

        # GH #423
        if len(frame._get_axis(axis)) == 0:
            result = Series(0, index=frame._get_agg_axis(axis))
        else:
            if frame._is_mixed_type:
                result = notnull(frame).sum(axis=axis)
            else:
                counts = notnull(frame.values).sum(axis=axis)
                result = Series(counts, index=frame._get_agg_axis(axis))

        return result.astype('int64')

    def _count_level(self, level, axis=0, numeric_only=False):
        if numeric_only:
            frame = self._get_numeric_data()
        else:
            frame = self

        count_axis = frame._get_axis(axis)
        agg_axis = frame._get_agg_axis(axis)

        if not isinstance(count_axis, MultiIndex):
            raise TypeError("Can only count levels on hierarchical %s." %
                            self._get_axis_name(axis))

        if frame._is_mixed_type:
            # Since we have mixed types, calling notnull(frame.values) might
            # upcast everything to object
            mask = notnull(frame).values
        else:
            # But use the speedup when we have homogeneous dtypes
            mask = notnull(frame.values)

        if axis == 1:
            # We're transposing the mask rather than frame to avoid potential
            # upcasts to object, which induces a ~20x slowdown
            mask = mask.T

        if isinstance(level, compat.string_types):
            level = count_axis._get_level_number(level)

        level_index = count_axis.levels[level]
        labels = com._ensure_int64(count_axis.labels[level])
        counts = lib.count_level_2d(mask, labels, len(level_index))

        result = DataFrame(counts, index=level_index,
                           columns=agg_axis)

        if axis == 1:
            # Undo our earlier transpose
            return result.T
        else:
            return result

    def _reduce(self, op, name, axis=0, skipna=True, numeric_only=None,
                filter_type=None, **kwds):
        axis = self._get_axis_number(axis)
        f = lambda x: op(x, axis=axis, skipna=skipna, **kwds)
        labels = self._get_agg_axis(axis)

        # exclude timedelta/datetime unless we are uniform types
        if axis == 1 and self._is_mixed_type and self._is_datelike_mixed_type:
            numeric_only = True

        if numeric_only is None:
            try:
                values = self.values
                result = f(values)
            except Exception as e:

                # try by-column first
                if filter_type is None and axis == 0:
                    try:

                        # this can end up with a non-reduction
                        # but not always. if the types are mixed
                        # with datelike then need to make sure a series
                        result = self.apply(f,reduce=False)
                        if result.ndim == self.ndim:
                            result = result.iloc[0]
                        return result
                    except:
                        pass

                if filter_type is None or filter_type == 'numeric':
                    data = self._get_numeric_data()
                elif filter_type == 'bool':
                    data = self._get_bool_data()
                else:  # pragma: no cover
                    e = NotImplementedError("Handling exception with filter_"
                                            "type %s not implemented."
                                            % filter_type)
                    raise_with_traceback(e)
                result = f(data.values)
                labels = data._get_agg_axis(axis)
        else:
            if numeric_only:
                if filter_type is None or filter_type == 'numeric':
                    data = self._get_numeric_data()
                elif filter_type == 'bool':
                    data = self._get_bool_data()
                else:  # pragma: no cover
                    msg = ("Generating numeric_only data with filter_type %s"
                           "not supported." % filter_type)
                    raise NotImplementedError(msg)
                values = data.values
                labels = data._get_agg_axis(axis)
            else:
                values = self.values
            result = f(values)

        if is_object_dtype(result.dtype):
            try:
                if filter_type is None or filter_type == 'numeric':
                    result = result.astype(np.float64)
                elif filter_type == 'bool' and notnull(result).all():
                    result = result.astype(np.bool_)
            except (ValueError, TypeError):

                # try to coerce to the original dtypes item by item if we can
                if axis == 0:
                    result = com._coerce_to_dtypes(result, self.dtypes)

        return Series(result, index=labels)

    def idxmin(self, axis=0, skipna=True):
        """
        Return index of first occurrence of minimum over requested axis.
        NA/null values are excluded.

        Parameters
        ----------
        axis : {0, 1}
            0 for row-wise, 1 for column-wise
        skipna : boolean, default True
            Exclude NA/null values. If an entire row/column is NA, the result
            will be NA

        Returns
        -------
        idxmin : Series

        Notes
        -----
        This method is the DataFrame version of ``ndarray.argmin``.

        See Also
        --------
        Series.idxmin
        """
        axis = self._get_axis_number(axis)
        indices = nanops.nanargmin(self.values, axis=axis, skipna=skipna)
        index = self._get_axis(axis)
        result = [index[i] if i >= 0 else NA for i in indices]
        return Series(result, index=self._get_agg_axis(axis))

    def idxmax(self, axis=0, skipna=True):
        """
        Return index of first occurrence of maximum over requested axis.
        NA/null values are excluded.

        Parameters
        ----------
        axis : {0, 1}
            0 for row-wise, 1 for column-wise
        skipna : boolean, default True
            Exclude NA/null values. If an entire row/column is NA, the result
            will be first index.

        Returns
        -------
        idxmax : Series

        Notes
        -----
        This method is the DataFrame version of ``ndarray.argmax``.

        See Also
        --------
        Series.idxmax
        """
        axis = self._get_axis_number(axis)
        indices = nanops.nanargmax(self.values, axis=axis, skipna=skipna)
        index = self._get_axis(axis)
        result = [index[i] if i >= 0 else NA for i in indices]
        return Series(result, index=self._get_agg_axis(axis))

    def _get_agg_axis(self, axis_num):
        """ let's be explict about this """
        if axis_num == 0:
            return self.columns
        elif axis_num == 1:
            return self.index
        else:
            raise ValueError('Axis must be 0 or 1 (got %r)' % axis_num)

    def mode(self, axis=0, numeric_only=False):
        """
        Gets the mode of each element along the axis selected. Empty if nothing
        has 2+ occurrences. Adds a row for each mode per label, fills in gaps
        with nan.

        Parameters
        ----------
        axis : {0, 1, 'index', 'columns'} (default 0)
            * 0/'index' : get mode of each column
            * 1/'columns' : get mode of each row
        numeric_only : boolean, default False
            if True, only apply to numeric columns

        Returns
        -------
        modes : DataFrame (sorted)
        """
        data = self if not numeric_only else self._get_numeric_data()
        f = lambda s: s.mode()
        return data.apply(f, axis=axis)

    def quantile(self, q=0.5, axis=0, numeric_only=True):
        """
        Return values at the given quantile over requested axis, a la
        numpy.percentile.

        Parameters
        ----------
        q : float or array-like, default 0.5 (50% quantile)
            0 <= q <= 1, the quantile(s) to compute
        axis : {0, 1, 'index', 'columns'} (default 0)
            0 or 'index' for row-wise, 1 or 'columns' for column-wise 


        Returns
        -------
        quantiles : Series or DataFrame
            If ``q`` is an array, a DataFrame will be returned where the
            index is ``q``, the columns are the columns of self, and the
            values are the quantiles.
            If ``q`` is a float, a Series will be returned where the
            index is the columns of self and the values are the quantiles.

        Examples
        --------

        >>> df = DataFrame(np.array([[1, 1], [2, 10], [3, 100], [4, 100]]),
                          columns=['a', 'b'])
        >>> df.quantile(.1)
        a    1.3
        b    3.7
        dtype: float64
        >>> df.quantile([.1, .5])
               a     b
        0.1  1.3   3.7
        0.5  2.5  55.0
        """
        per = np.asarray(q) * 100

        if not com.is_list_like(per):
            per = [per]
            q = [q]
            squeeze = True
        else:
            squeeze = False

        def f(arr, per):
            if arr._is_datelike_mixed_type:
                values = _values_from_object(arr).view('i8')
            else:
                values = arr.astype(float)
            values = values[notnull(values)]
            if len(values) == 0:
                return NA
            else:
                return _quantile(values, per)

        data = self._get_numeric_data() if numeric_only else self

        axis = self._get_axis_number(axis)

        if axis == 1:
            data = data.T

        # need to know which cols are timestamp going in so that we can
        # map timestamp over them after getting the quantile.
        is_dt_col = data.dtypes.map(com.is_datetime64_dtype)
        is_dt_col = is_dt_col[is_dt_col].index

        quantiles = [[f(vals, x) for x in per]
                     for (_, vals) in data.iteritems()]
        result = DataFrame(quantiles, index=data._info_axis, columns=q).T
        if len(is_dt_col) > 0:
            result[is_dt_col] = result[is_dt_col].applymap(lib.Timestamp)
        if squeeze:
            if result.shape == (1, 1):
                result = result.T.iloc[:, 0]  # don't want scalar
            else:
                result = result.T.squeeze()
            result.name = None  # For groupby, so it can set an index name
        return result

    def rank(self, axis=0, numeric_only=None, method='average',
             na_option='keep', ascending=True, pct=False):
        """
        Compute numerical data ranks (1 through n) along axis. Equal values are
        assigned a rank that is the average of the ranks of those values

        Parameters
        ----------
        axis : {0, 1}, default 0
            Ranks over columns (0) or rows (1)
        numeric_only : boolean, default None
            Include only float, int, boolean data
        method : {'average', 'min', 'max', 'first', 'dense'}
            * average: average rank of group
            * min: lowest rank in group
            * max: highest rank in group
            * first: ranks assigned in order they appear in the array
            * dense: like 'min', but rank always increases by 1 between groups
        na_option : {'keep', 'top', 'bottom'}
            * keep: leave NA values where they are
            * top: smallest rank if ascending
            * bottom: smallest rank if descending
        ascending : boolean, default True
            False for ranks by high (1) to low (N)
        pct : boolean, default False
            Computes percentage rank of data

        Returns
        -------
        ranks : DataFrame
        """
        axis = self._get_axis_number(axis)
        if numeric_only is None:
            try:
                ranks = algos.rank(self.values, axis=axis, method=method,
                                   ascending=ascending, na_option=na_option,
                                   pct=pct)
                return self._constructor(ranks, index=self.index,
                                         columns=self.columns)
            except TypeError:
                numeric_only = True
        if numeric_only:
            data = self._get_numeric_data()
        else:
            data = self
        ranks = algos.rank(data.values, axis=axis, method=method,
                           ascending=ascending, na_option=na_option, pct=pct)
        return self._constructor(ranks, index=data.index, columns=data.columns)

    def to_timestamp(self, freq=None, how='start', axis=0, copy=True):
        """
        Cast to DatetimeIndex of timestamps, at *beginning* of period

        Parameters
        ----------
        freq : string, default frequency of PeriodIndex
            Desired frequency
        how : {'s', 'e', 'start', 'end'}
            Convention for converting period to timestamp; start of period
            vs. end
        axis : {0, 1} default 0
            The axis to convert (the index by default)
        copy : boolean, default True
            If false then underlying input data is not copied

        Returns
        -------
        df : DataFrame with DatetimeIndex
        """
        new_data = self._data
        if copy:
            new_data = new_data.copy()

        axis = self._get_axis_number(axis)
        if axis == 0:
            new_data.set_axis(1, self.index.to_timestamp(freq=freq, how=how))
        elif axis == 1:
            new_data.set_axis(0, self.columns.to_timestamp(freq=freq, how=how))
        else:  # pragma: no cover
            raise AssertionError('Axis must be 0 or 1. Got %s' % str(axis))

        return self._constructor(new_data)

    def to_period(self, freq=None, axis=0, copy=True):
        """
        Convert DataFrame from DatetimeIndex to PeriodIndex with desired
        frequency (inferred from index if not passed)

        Parameters
        ----------
        freq : string, default
        axis : {0, 1}, default 0
            The axis to convert (the index by default)
        copy : boolean, default True
            If False then underlying input data is not copied

        Returns
        -------
        ts : TimeSeries with PeriodIndex
        """
        new_data = self._data
        if copy:
            new_data = new_data.copy()

        axis = self._get_axis_number(axis)
        if axis == 0:
            new_data.set_axis(1, self.index.to_period(freq=freq))
        elif axis == 1:
            new_data.set_axis(0, self.columns.to_period(freq=freq))
        else:  # pragma: no cover
            raise AssertionError('Axis must be 0 or 1. Got %s' % str(axis))

        return self._constructor(new_data)

    def isin(self, values):
        """
        Return boolean DataFrame showing whether each element in the
        DataFrame is contained in values.

        Parameters
        ----------
        values : iterable, Series, DataFrame or dictionary
            The result will only be true at a location if all the
            labels match. If `values` is a Series, that's the index. If
            `values` is a dictionary, the keys must be the column names,
            which must match. If `values` is a DataFrame,
            then both the index and column labels must match.

        Returns
        -------

        DataFrame of booleans

        Examples
        --------
        When ``values`` is a list:

        >>> df = DataFrame({'A': [1, 2, 3], 'B': ['a', 'b', 'f']})
        >>> df.isin([1, 3, 12, 'a'])
               A      B
        0   True   True
        1  False  False
        2   True  False

        When ``values`` is a dict:

        >>> df = DataFrame({'A': [1, 2, 3], 'B': [1, 4, 7]})
        >>> df.isin({'A': [1, 3], 'B': [4, 7, 12]})
               A      B
        0   True  False  # Note that B didn't match the 1 here.
        1  False   True
        2   True   True

        When ``values`` is a Series or DataFrame:

        >>> df = DataFrame({'A': [1, 2, 3], 'B': ['a', 'b', 'f']})
        >>> other = DataFrame({'A': [1, 3, 3, 2], 'B': ['e', 'f', 'f', 'e']})
        >>> df.isin(other)
               A      B
        0   True  False
        1  False  False  # Column A in `other` has a 3, but not at index 1.
        2   True   True
        """
        if isinstance(values, dict):
            from collections import defaultdict
            from pandas.tools.merge import concat
            values = defaultdict(list, values)
            return concat((self.iloc[:, [i]].isin(values[col])
                           for i, col in enumerate(self.columns)), axis=1)
        elif isinstance(values, Series):
            if not values.index.is_unique:
                raise ValueError("ValueError: cannot compute isin with"
                                 " a duplicate axis.")
            return self.eq(values.reindex_like(self), axis='index')
        elif isinstance(values, DataFrame):
            if not (values.columns.is_unique and values.index.is_unique):
                raise ValueError("ValueError: cannot compute isin with"
                                 " a duplicate axis.")
            return self.eq(values.reindex_like(self))
        else:
            if not is_list_like(values):
                raise TypeError("only list-like or dict-like objects are"
                                " allowed to be passed to DataFrame.isin(), "
                                "you passed a "
                                "{0!r}".format(type(values).__name__))
            return DataFrame(lib.ismember(self.values.ravel(),
                                          set(values)).reshape(self.shape),
                             self.index,
                             self.columns)

    #----------------------------------------------------------------------
    # Deprecated stuff

    def combineAdd(self, other):
        """
        Add two DataFrame objects and do not propagate
        NaN values, so if for a (column, time) one frame is missing a
        value, it will default to the other frame's value (which might
        be NaN as well)

        Parameters
        ----------
        other : DataFrame

        Returns
        -------
        DataFrame
        """
        return self.add(other, fill_value=0.)

    def combineMult(self, other):
        """
        Multiply two DataFrame objects and do not propagate NaN values, so if
        for a (column, time) one frame is missing a value, it will default to
        the other frame's value (which might be NaN as well)

        Parameters
        ----------
        other : DataFrame

        Returns
        -------
        DataFrame
        """
        return self.mul(other, fill_value=1.)


DataFrame._setup_axes(['index', 'columns'], info_axis=1, stat_axis=0,
                      axes_are_reversed=True, aliases={'rows': 0})
DataFrame._add_numeric_operations()

_EMPTY_SERIES = Series([])

def _arrays_to_mgr(arrays, arr_names, index, columns, dtype=None):
    """
    Segregate Series based on type and coerce into matrices.
    Needs to handle a lot of exceptional cases.
    """
    # figure out the index, if necessary
    if index is None:
        index = extract_index(arrays)
    else:
        index = _ensure_index(index)

    # don't force copy because getting jammed in an ndarray anyway
    arrays = _homogenize(arrays, index, dtype)

    # from BlockManager perspective
    axes = [_ensure_index(columns), _ensure_index(index)]

    return create_block_manager_from_arrays(arrays, arr_names, axes)


def extract_index(data):
    from pandas.core.index import _union_indexes

    index = None
    if len(data) == 0:
        index = Index([])
    elif len(data) > 0:
        raw_lengths = []
        indexes = []

        have_raw_arrays = False
        have_series = False
        have_dicts = False

        for v in data:
            if isinstance(v, Series):
                have_series = True
                indexes.append(v.index)
            elif isinstance(v, dict):
                have_dicts = True
                indexes.append(list(v.keys()))
            elif is_list_like(v) and getattr(v, 'ndim', 1) == 1:
                have_raw_arrays = True
                raw_lengths.append(len(v))

        if not indexes and not raw_lengths:
            raise ValueError('If using all scalar values, you must pass'
                             ' an index')

        if have_series or have_dicts:
            index = _union_indexes(indexes)

        if have_raw_arrays:
            lengths = list(set(raw_lengths))
            if len(lengths) > 1:
                raise ValueError('arrays must all be same length')

            if have_dicts:
                raise ValueError('Mixing dicts with non-Series may lead to '
                                 'ambiguous ordering.')

            if have_series:
                if lengths[0] != len(index):
                    msg = ('array length %d does not match index length %d'
                           % (lengths[0], len(index)))
                    raise ValueError(msg)
            else:
                index = Index(np.arange(lengths[0]))

    return _ensure_index(index)


def _prep_ndarray(values, copy=True):
    if not isinstance(values, (np.ndarray, Series, Index)):
        if len(values) == 0:
            return np.empty((0, 0), dtype=object)

        def convert(v):
            return com._possibly_convert_platform(v)

        # we could have a 1-dim or 2-dim list here
        # this is equiv of np.asarray, but does object conversion
        # and platform dtype preservation
        try:
            if com.is_list_like(values[0]) or hasattr(values[0], 'len'):
                values = np.array([convert(v) for v in values])
            else:
                values = convert(values)
        except:
            values = convert(values)

    else:

        # drop subclass info, do not copy data
        values = np.asarray(values)
        if copy:
            values = values.copy()

    if values.ndim == 1:
        values = values.reshape((values.shape[0], 1))
    elif values.ndim != 2:
        raise ValueError('Must pass 2-d input')

    return values


def _to_arrays(data, columns, coerce_float=False, dtype=None):
    """
    Return list of arrays, columns
    """
    if isinstance(data, DataFrame):
        if columns is not None:
            arrays = [data.icol(i).values for i, col in enumerate(data.columns)
                      if col in columns]
        else:
            columns = data.columns
            arrays = [data.icol(i).values for i in range(len(columns))]

        return arrays, columns

    if not len(data):
        if isinstance(data, np.ndarray):
            columns = data.dtype.names
            if columns is not None:
                return [[]] * len(columns), columns
        return [], []  # columns if columns is not None else []
    if isinstance(data[0], (list, tuple)):
        return _list_to_arrays(data, columns, coerce_float=coerce_float,
                               dtype=dtype)
    elif isinstance(data[0], collections.Mapping):
        return _list_of_dict_to_arrays(data, columns,
                                       coerce_float=coerce_float,
                                       dtype=dtype)
    elif isinstance(data[0], Series):
        return _list_of_series_to_arrays(data, columns,
                                         coerce_float=coerce_float,
                                         dtype=dtype)
    elif isinstance(data[0], Categorical):
        if columns is None:
            columns = _default_index(len(data))
        return data, columns
    elif (isinstance(data, (np.ndarray, Series, Index))
          and data.dtype.names is not None):

        columns = list(data.dtype.names)
        arrays = [data[k] for k in columns]
        return arrays, columns
    else:
        # last ditch effort
        data = lmap(tuple, data)
        return _list_to_arrays(data, columns,
                               coerce_float=coerce_float,
                               dtype=dtype)


def _masked_rec_array_to_mgr(data, index, columns, dtype, copy):
    """ extract from a masked rec array and create the manager """

    # essentially process a record array then fill it
    fill_value = data.fill_value
    fdata = ma.getdata(data)
    if index is None:
        index = _get_names_from_index(fdata)
        if index is None:
            index = _default_index(len(data))
    index = _ensure_index(index)

    if columns is not None:
        columns = _ensure_index(columns)
    arrays, arr_columns = _to_arrays(fdata, columns)

    # fill if needed
    new_arrays = []
    for fv, arr, col in zip(fill_value, arrays, arr_columns):
        mask = ma.getmaskarray(data[col])
        if mask.any():
            arr, fv = _maybe_upcast(arr, fill_value=fv, copy=True)
            arr[mask] = fv
        new_arrays.append(arr)

    # create the manager
    arrays, arr_columns = _reorder_arrays(new_arrays, arr_columns, columns)
    if columns is None:
        columns = arr_columns

    mgr = _arrays_to_mgr(arrays, arr_columns, index, columns)

    if copy:
        mgr = mgr.copy()
    return mgr


def _reorder_arrays(arrays, arr_columns, columns):
    # reorder according to the columns
    if (columns is not None and len(columns) and arr_columns is not None and
            len(arr_columns)):
        indexer = _ensure_index(
            arr_columns).get_indexer(columns)
        arr_columns = _ensure_index(
            [arr_columns[i] for i in indexer])
        arrays = [arrays[i] for i in indexer]
    return arrays, arr_columns


def _list_to_arrays(data, columns, coerce_float=False, dtype=None):
    if len(data) > 0 and isinstance(data[0], tuple):
        content = list(lib.to_object_array_tuples(data).T)
    else:
        # list of lists
        content = list(lib.to_object_array(data).T)
    return _convert_object_array(content, columns, dtype=dtype,
                                 coerce_float=coerce_float)


def _list_of_series_to_arrays(data, columns, coerce_float=False, dtype=None):
    from pandas.core.index import _get_combined_index

    if columns is None:
        columns = _get_combined_index([
            s.index for s in data if getattr(s, 'index', None) is not None
        ])

    indexer_cache = {}

    aligned_values = []
    for s in data:
        index = getattr(s, 'index', None)
        if index is None:
            index = _default_index(len(s))

        if id(index) in indexer_cache:
            indexer = indexer_cache[id(index)]
        else:
            indexer = indexer_cache[id(index)] = index.get_indexer(columns)

        values = _values_from_object(s)
        aligned_values.append(com.take_1d(values, indexer))

    values = np.vstack(aligned_values)

    if values.dtype == np.object_:
        content = list(values.T)
        return _convert_object_array(content, columns, dtype=dtype,
                                     coerce_float=coerce_float)
    else:
        return values.T, columns


def _list_of_dict_to_arrays(data, columns, coerce_float=False, dtype=None):
    if columns is None:
        gen = (list(x.keys()) for x in data)
        columns = lib.fast_unique_multiple_list_gen(gen)

    # assure that they are of the base dict class and not of derived
    # classes
    data = [(type(d) is dict) and d or dict(d) for d in data]

    content = list(lib.dicts_to_array(data, list(columns)).T)
    return _convert_object_array(content, columns, dtype=dtype,
                                 coerce_float=coerce_float)


def _convert_object_array(content, columns, coerce_float=False, dtype=None):
    if columns is None:
        columns = _default_index(len(content))
    else:
        if len(columns) != len(content):  # pragma: no cover
            # caller's responsibility to check for this...
            raise AssertionError('%d columns passed, passed data had %s '
                                 'columns' % (len(columns), len(content)))

    # provide soft conversion of object dtypes
    def convert(arr):
        if dtype != object and dtype != np.object:
            arr = lib.maybe_convert_objects(arr, try_float=coerce_float)
            arr = com._possibly_cast_to_datetime(arr, dtype)
        return arr

    arrays = [ convert(arr) for arr in content ]

    return arrays, columns


def _get_names_from_index(data):
    index = lrange(len(data))
    has_some_name = any([getattr(s, 'name', None) is not None for s in data])
    if not has_some_name:
        return index

    count = 0
    for i, s in enumerate(data):
        n = getattr(s, 'name', None)
        if n is not None:
            index[i] = n
        else:
            index[i] = 'Unnamed %d' % count
            count += 1

    return index


def _homogenize(data, index, dtype=None):
    from pandas.core.series import _sanitize_array

    oindex = None
    homogenized = []

    for v in data:
        if isinstance(v, Series):
            if dtype is not None:
                v = v.astype(dtype)
            if v.index is not index:
                # Forces alignment. No need to copy data since we
                # are putting it into an ndarray later
                v = v.reindex(index, copy=False)
        else:
            if isinstance(v, dict):
                if oindex is None:
                    oindex = index.astype('O')
                if type(v) == dict:
                    # fast cython method
                    v = lib.fast_multiget(v, oindex.values, default=NA)
                else:
                    v = lib.map_infer(oindex.values, v.get)

            v = _sanitize_array(v, index, dtype=dtype, copy=False,
                                raise_cast_failure=False)

        homogenized.append(v)

    return homogenized


def _from_nested_dict(data):
    # TODO: this should be seriously cythonized
    new_data = OrderedDict()
    for index, s in compat.iteritems(data):
        for col, v in compat.iteritems(s):
            new_data[col] = new_data.get(col, OrderedDict())
            new_data[col][index] = v
    return new_data


def _put_str(s, space):
    return ('%s' % s)[:space].ljust(space)


#----------------------------------------------------------------------
# Add plotting methods to DataFrame

import pandas.tools.plotting as gfx

DataFrame.plot = gfx.plot_frame
DataFrame.hist = gfx.hist_frame


@Appender(_shared_docs['boxplot'] % _shared_doc_kwargs)
def boxplot(self, column=None, by=None, ax=None, fontsize=None,
            rot=0, grid=True, figsize=None, layout=None, return_type=None,
            **kwds):
    import pandas.tools.plotting as plots
    import matplotlib.pyplot as plt
    ax = plots.boxplot(self, column=column, by=by, ax=ax,
                       fontsize=fontsize, grid=grid, rot=rot,
                       figsize=figsize, layout=layout, return_type=return_type,
                       **kwds)
    plt.draw_if_interactive()
    return ax

DataFrame.boxplot = boxplot

ops.add_flex_arithmetic_methods(DataFrame, **ops.frame_flex_funcs)
ops.add_special_arithmetic_methods(DataFrame, **ops.frame_special_funcs)

if __name__ == '__main__':
    import nose
    nose.runmodule(argv=[__file__, '-vvs', '-x', '--pdb', '--pdb-failure'],
                   exit=False)