pytables.py

"""
High level interface to PyTables for reading and writing pandas data structures
to disk
"""

# pylint: disable-msg=E1101,W0613,W0603

from datetime import datetime, date
import time
import re

import numpy as np
from pandas import (
    Series, TimeSeries, DataFrame, Panel, Index, MultiIndex, Int64Index
)
from pandas.sparse.api import SparseSeries, SparseDataFrame, SparsePanel
from pandas.sparse.array import BlockIndex, IntIndex
from pandas.tseries.api import PeriodIndex, DatetimeIndex
from pandas.core.common import adjoin
from pandas.core.algorithms import match, unique

from pandas.core.categorical import Factor
from pandas.core.common import _asarray_tuplesafe
from pandas.core.internals import BlockManager, make_block
from pandas.core.reshape import block2d_to_block3d
import pandas.core.common as com

import pandas.lib as lib
from contextlib import contextmanager

# reading and writing the full object in one go
_TYPE_MAP = {
    Series: 'series',
    SparseSeries: 'sparse_series',
    TimeSeries: 'series',
    DataFrame: 'frame',
    SparseDataFrame: 'sparse_frame',
    Panel: 'wide',
    SparsePanel: 'sparse_panel'
}

_NAME_MAP = {
    'series': 'Series',
    'time_series': 'TimeSeries',
    'sparse_series': 'SparseSeries',
    'frame': 'DataFrame',
    'sparse_frame': 'SparseDataFrame',
    'frame_table': 'DataFrame (Table)',
    'wide': 'Panel',
    'sparse_panel': 'SparsePanel',
    'wide_table': 'Panel (Table)',
    'long': 'LongPanel',
    # legacy h5 files
    'Series': 'Series',
    'TimeSeries': 'TimeSeries',
    'DataFrame': 'DataFrame',
    'DataMatrix': 'DataMatrix'
}

# legacy handlers
_LEGACY_MAP = {
    'Series': 'legacy_series',
    'TimeSeries': 'legacy_series',
    'DataFrame': 'legacy_frame',
    'DataMatrix': 'legacy_frame',
    'WidePanel': 'wide_table',
}

# oh the troubles to reduce import time
_table_mod = None

def _tables():
    global _table_mod
    if _table_mod is None:
        import tables
        _table_mod = tables
    return _table_mod


@contextmanager
def get_store(path, mode='a', complevel=None, complib=None,
              fletcher32=False):
    """
    Creates an HDFStore instance. This function can be used in a with statement

    Parameters
    ----------
    path : string
        File path to HDF5 file
    mode : {'a', 'w', 'r', 'r+'}, default 'a'

        ``'r'``
            Read-only; no data can be modified.
        ``'w'``
            Write; a new file is created (an existing file with the same
            name would be deleted).
        ``'a'``
            Append; an existing file is opened for reading and writing,
            and if the file does not exist it is created.
        ``'r+'``
            It is similar to ``'a'``, but the file must already exist.
    complevel : int, 1-9, default 0
            If a complib is specified compression will be applied
            where possible
    complib : {'zlib', 'bzip2', 'lzo', 'blosc', None}, default None
            If complevel is > 0 apply compression to objects written
            in the store wherever possible
    fletcher32 : bool, default False
            If applying compression use the fletcher32 checksum

    Examples
    --------
    >>> with get_store('test.h5') as store:
    >>>     store['foo'] = bar   # write to HDF5
    >>>     bar = store['foo']   # retrieve
    """
    store = None
    try:
        store = HDFStore(path, mode=mode, complevel=complevel,
                         complib=complib, fletcher32=False)
        yield store
    finally:
        if store is not None:
            store.close()


class HDFStore(object):
    """
    dict-like IO interface for storing pandas objects in PyTables
    format.

    DataFrame and Panel can be stored in Table format, which is slower to
    read and write but can be searched and manipulated more like an SQL
    table. See HDFStore.put for more information

    Parameters
    ----------
    path : string
        File path to HDF5 file
    mode : {'a', 'w', 'r', 'r+'}, default 'a'

        ``'r'``
            Read-only; no data can be modified.
        ``'w'``
            Write; a new file is created (an existing file with the same
            name would be deleted).
        ``'a'``
            Append; an existing file is opened for reading and writing,
            and if the file does not exist it is created.
        ``'r+'``
            It is similar to ``'a'``, but the file must already exist.
    complevel : int, 1-9, default 0
            If a complib is specified compression will be applied
            where possible
    complib : {'zlib', 'bzip2', 'lzo', 'blosc', None}, default None
            If complevel is > 0 apply compression to objects written
            in the store wherever possible
    fletcher32 : bool, default False
            If applying compression use the fletcher32 checksum

    Examples
    --------
    >>> store = HDFStore('test.h5')
    >>> store['foo'] = bar   # write to HDF5
    >>> bar = store['foo']   # retrieve
    >>> store.close()
    """
    _quiet = False

    def __init__(self, path, mode='a', complevel=None, complib=None,
                 fletcher32=False):
        try:
            import tables as _
        except ImportError:  # pragma: no cover
            raise Exception('HDFStore requires PyTables')

        self.path = path
        self.mode = mode
        self.handle = None
        self.complevel = complevel
        self.complib = complib
        self.fletcher32 = fletcher32
        self.filters = None
        self.open(mode=mode, warn=False)

    def __getitem__(self, key):
        return self.get(key)

    def __setitem__(self, key, value):
        self.put(key, value)

    def __contains__(self, key):
        return hasattr(self.handle.root, key)

    def __len__(self):
        return len(self.handle.root._v_children)

    def __repr__(self):
        output = '%s\nFile path: %s\n' % (type(self), self.path)

        if len(self) > 0:
            keys = []
            values = []
            for k, v in sorted(self.handle.root._v_children.iteritems()):
                kind = v._v_attrs.pandas_type

                keys.append(str(k))
                values.append(_NAME_MAP[kind])

            output += adjoin(5, keys, values)
        else:
            output += 'Empty'

        return output

    def keys(self):
        """
        Return a (potentially unordered) list of the keys corresponding to the
        objects stored in the HDFStore
        """
        return self.handle.root._v_children.keys()

    def open(self, mode='a', warn=True):
        """
        Open the file in the specified mode

        Parameters
        ----------
        mode : {'a', 'w', 'r', 'r+'}, default 'a'
            See HDFStore docstring or tables.openFile for info about modes
        """
        self.mode = mode
        if warn and mode == 'w':  # pragma: no cover
            while True:
                response = raw_input("Re-opening as mode='w' will delete the "
                                     "current file. Continue (y/n)?")
                if response == 'y':
                    break
                elif response == 'n':
                    return
        if self.handle is not None and self.handle.isopen:
            self.handle.close()

        if self.complib is not None:
            if self.complevel is None:
                self.complevel = 9
            self.filters = _tables().Filters(self.complevel,
                                             self.complib,
                                             fletcher32=self.fletcher32)

        try:
            self.handle = _tables().openFile(self.path, self.mode)
        except IOError, e:  # pragma: no cover
            if 'can not be written' in str(e):
                print 'Opening %s in read-only mode' % self.path
                self.handle = _tables().openFile(self.path, 'r')
            else:
                raise

    def close(self):
        """
        Close the PyTables file handle
        """
        self.handle.close()

    def flush(self):
        """
        Force all buffered modifications to be written to disk
        """
        self.handle.flush()

    def get(self, key):
        """
        Retrieve pandas object stored in file

        Parameters
        ----------
        key : object

        Returns
        -------
        obj : type of object stored in file
        """
        exc_type = _tables().NoSuchNodeError
        try:
            group = getattr(self.handle.root, key)
            return self._read_group(group)
        except (exc_type, AttributeError):
            raise KeyError('No object named %s in the file' % key)

    def select(self, key, where=None):
        """
        Retrieve pandas object stored in file, optionally based on where
        criteria

        Parameters
        ----------
        key : object
        where : list, optional

           Must be a list of dict objects of the following forms. Selection can
           be performed on the 'index' or 'column' fields.

           Comparison op
               {'field' : 'index',
                'op'    : '>=',
                'value' : value}

           Match single value
               {'field' : 'index',
                'value' : v1}

           Match a set of values
               {'field' : 'index',
                'value' : [v1, v2, v3]}

        """
        group = getattr(self.handle.root, key, None)
        if 'table' not in group._v_attrs.pandas_type:
            raise Exception('can only select on objects written as tables')
        if group is not None:
            return self._read_group(group, where)

    def put(self, key, value, table=False, append=False,
            compression=None):
        """
        Store object in HDFStore

        Parameters
        ----------
        key : object
        value : {Series, DataFrame, Panel}
        table : boolean, default False
            Write as a PyTables Table structure which may perform worse but
            allow more flexible operations like searching / selecting subsets
            of the data
        append : boolean, default False
            For table data structures, append the input data to the existing
            table
        compression : {None, 'blosc', 'lzo', 'zlib'}, default None
            Use a compression algorithm to compress the data
            If None, the compression settings specified in the ctor will
            be used.
        """
        self._write_to_group(key, value, table=table, append=append,
                             comp=compression)

    def _get_handler(self, op, kind):
        return getattr(self, '_%s_%s' % (op, kind))

    def remove(self, key, where=None):
        """
        Remove pandas object partially by specifying the where condition

        Parameters
        ----------
        key : string
            Node to remove or delete rows from
        where : list
            For Table node, delete specified rows. See HDFStore.select for more
            information

        Parameters
        ----------
        key : object
        """
        if where is None:
            self.handle.removeNode(self.handle.root, key, recursive=True)
        else:
            group = getattr(self.handle.root, key, None)
            if group is not None:
                self._delete_from_table(group, where)

    def append(self, key, value):
        """
        Append to Table in file. Node must already exist and be Table
        format.

        Parameters
        ----------
        key : object
        value : {Series, DataFrame, Panel}

        Notes
        -----
        Does *not* check if data being appended overlaps with existing
        data in the table, so be careful
        """
        self._write_to_group(key, value, table=True, append=True)

    def _write_to_group(self, key, value, table=False, append=False,
                        comp=None):
        root = self.handle.root
        if key not in root._v_children:
            group = self.handle.createGroup(root, key)
        else:
            group = getattr(root, key)

        kind = _TYPE_MAP[type(value)]
        if table or (append and _is_table_type(group)):
            kind = '%s_table' % kind
            handler = self._get_handler(op='write', kind=kind)
            wrapper = lambda value: handler(group, value, append=append,
                                            comp=comp)
        else:
            if append:
                raise ValueError('Can only append to Tables')
            if comp:
                raise ValueError('Compression only supported on Tables')

            handler = self._get_handler(op='write', kind=kind)
            wrapper = lambda value: handler(group, value)

        wrapper(value)
        group._v_attrs.pandas_type = kind

    def _write_series(self, group, series):
        self._write_index(group, 'index', series.index)
        self._write_array(group, 'values', series.values)
        group._v_attrs.name = series.name

    def _write_sparse_series(self, group, series):
        self._write_index(group, 'index', series.index)
        self._write_index(group, 'sp_index', series.sp_index)
        self._write_array(group, 'sp_values', series.sp_values)
        group._v_attrs.name = series.name
        group._v_attrs.fill_value = series.fill_value
        group._v_attrs.kind = series.kind

    def _read_sparse_series(self, group, where=None):
        index = self._read_index(group, 'index')
        sp_values = _read_array(group, 'sp_values')
        sp_index = self._read_index(group, 'sp_index')
        name = getattr(group._v_attrs, 'name', None)
        fill_value = getattr(group._v_attrs, 'fill_value', None)
        kind = getattr(group._v_attrs, 'kind', 'block')
        return SparseSeries(sp_values, index=index, sparse_index=sp_index,
                            kind=kind, fill_value=fill_value,
                            name=name)

    def _write_sparse_frame(self, group, sdf):
        for name, ss in sdf.iteritems():
            key = 'sparse_series_%s' % name
            if key not in group._v_children:
                node = self.handle.createGroup(group, key)
            else:
                node = getattr(group, key)
            self._write_sparse_series(node, ss)
        setattr(group._v_attrs, 'default_fill_value',
                sdf.default_fill_value)
        setattr(group._v_attrs, 'default_kind',
                sdf.default_kind)
        self._write_index(group, 'columns', sdf.columns)

    def _read_sparse_frame(self, group, where=None):
        columns = self._read_index(group, 'columns')
        sdict = {}
        for c in columns:
            key = 'sparse_series_%s' % c
            node = getattr(group, key)
            sdict[c] = self._read_sparse_series(node)
        default_kind = getattr(group._v_attrs, 'default_kind')
        default_fill_value = getattr(group._v_attrs, 'default_fill_value')
        return SparseDataFrame(sdict, columns=columns,
                               default_kind=default_kind,
                               default_fill_value=default_fill_value)

    def _write_sparse_panel(self, group, swide):
        setattr(group._v_attrs, 'default_fill_value', swide.default_fill_value)
        setattr(group._v_attrs, 'default_kind', swide.default_kind)
        self._write_index(group, 'items', swide.items)

        for name, sdf in swide.iteritems():
            key = 'sparse_frame_%s' % name
            if key not in group._v_children:
                node = self.handle.createGroup(group, key)
            else:
                node = getattr(group, key)
            self._write_sparse_frame(node, sdf)

    def _read_sparse_panel(self, group, where=None):
        default_fill_value = getattr(group._v_attrs, 'default_fill_value')
        default_kind = getattr(group._v_attrs, 'default_kind')
        items = self._read_index(group, 'items')

        sdict = {}
        for name in items:
            key = 'sparse_frame_%s' % name
            node = getattr(group, key)
            sdict[name] = self._read_sparse_frame(node)
        return SparsePanel(sdict, items=items, default_kind=default_kind,
                           default_fill_value=default_fill_value)

    def _write_frame(self, group, df):
        self._write_block_manager(group, df._data)

    def _read_frame(self, group, where=None):
        return DataFrame(self._read_block_manager(group))

    def _write_block_manager(self, group, data):
        if not data.is_consolidated():
            data = data.consolidate()

        group._v_attrs.ndim = data.ndim
        for i, ax in enumerate(data.axes):
            self._write_index(group, 'axis%d' % i, ax)

        # Supporting mixed-type DataFrame objects...nontrivial
        nblocks = len(data.blocks)
        group._v_attrs.nblocks = nblocks
        for i in range(nblocks):
            blk = data.blocks[i]
            self._write_index(group, 'block%d_items' % i, blk.items)
            self._write_array(group, 'block%d_values' % i, blk.values)

    def _read_block_manager(self, group):
        ndim = group._v_attrs.ndim

        axes = []
        for i in xrange(ndim):
            ax = self._read_index(group, 'axis%d' % i)
            axes.append(ax)

        items = axes[0]
        blocks = []
        for i in range(group._v_attrs.nblocks):
            blk_items = self._read_index(group, 'block%d_items' % i)
            values = _read_array(group, 'block%d_values' % i)
            blk = make_block(values, blk_items, items)
            blocks.append(blk)

        return BlockManager(blocks, axes)

    def _write_frame_table(self, group, df, append=False, comp=None):
        mat = df.values
        values = mat.reshape((1,) + mat.shape)

        if df._is_mixed_type:
            raise Exception('Cannot currently store mixed-type DataFrame '
                            'objects in Table format')

        self._write_table(group, items=['value'],
                          index=df.index, columns=df.columns,
                          values=values, append=append, compression=comp)

    def _write_wide(self, group, panel):
        panel._consolidate_inplace()
        self._write_block_manager(group, panel._data)

    def _read_wide(self, group, where=None):
        return Panel(self._read_block_manager(group))

    def _write_wide_table(self, group, panel, append=False, comp=None):
        self._write_table(group, items=panel.items, index=panel.major_axis,
                          columns=panel.minor_axis, values=panel.values,
                          append=append, compression=comp)

    def _read_wide_table(self, group, where=None):
        return self._read_panel_table(group, where)

    def _write_index(self, group, key, index):
        if isinstance(index, MultiIndex):
            setattr(group._v_attrs, '%s_variety' % key, 'multi')
            self._write_multi_index(group, key, index)
        elif isinstance(index, BlockIndex):
            setattr(group._v_attrs, '%s_variety' % key, 'block')
            self._write_block_index(group, key, index)
        elif isinstance(index, IntIndex):
            setattr(group._v_attrs, '%s_variety' % key, 'sparseint')
            self._write_sparse_intindex(group, key, index)
        else:
            setattr(group._v_attrs, '%s_variety' % key, 'regular')
            converted, kind, _ = _convert_index(index)
            self._write_array(group, key, converted)
            node = getattr(group, key)
            node._v_attrs.kind = kind
            node._v_attrs.name = index.name

            if isinstance(index, (DatetimeIndex, PeriodIndex)):
                node._v_attrs.index_class = _class_to_alias(type(index))

            if hasattr(index, 'freq'):
                node._v_attrs.freq = index.freq

            if hasattr(index, 'tz') and index.tz is not None:
                zone = lib.get_timezone(index.tz)
                if zone is None:
                    zone = lib.tot_seconds(index.tz.utcoffset())
                node._v_attrs.tz = zone

    def _read_index(self, group, key):
        variety = getattr(group._v_attrs, '%s_variety' % key)

        if variety == 'multi':
            return self._read_multi_index(group, key)
        elif variety == 'block':
            return self._read_block_index(group, key)
        elif variety == 'sparseint':
            return self._read_sparse_intindex(group, key)
        elif variety == 'regular':
            _, index = self._read_index_node(getattr(group, key))
            return index
        else:  # pragma: no cover
            raise Exception('unrecognized index variety: %s' % variety)

    def _write_block_index(self, group, key, index):
        self._write_array(group, '%s_blocs' % key, index.blocs)
        self._write_array(group, '%s_blengths' % key, index.blengths)
        setattr(group._v_attrs, '%s_length' % key, index.length)

    def _read_block_index(self, group, key):
        length = getattr(group._v_attrs, '%s_length' % key)
        blocs = _read_array(group, '%s_blocs' % key)
        blengths = _read_array(group, '%s_blengths' % key)
        return BlockIndex(length, blocs, blengths)

    def _write_sparse_intindex(self, group, key, index):
        self._write_array(group, '%s_indices' % key, index.indices)
        setattr(group._v_attrs, '%s_length' % key, index.length)

    def _read_sparse_intindex(self, group, key):
        length = getattr(group._v_attrs, '%s_length' % key)
        indices = _read_array(group, '%s_indices' % key)
        return IntIndex(length, indices)

    def _write_multi_index(self, group, key, index):
        setattr(group._v_attrs, '%s_nlevels' % key, index.nlevels)

        for i, (lev, lab, name) in enumerate(zip(index.levels,
                                                 index.labels,
                                                 index.names)):
            # write the level
            conv_level, kind, _ = _convert_index(lev)
            level_key = '%s_level%d' % (key, i)
            self._write_array(group, level_key, conv_level)
            node = getattr(group, level_key)
            node._v_attrs.kind = kind
            node._v_attrs.name = name

            # write the name
            setattr(node._v_attrs, '%s_name%d' % (key, i), name)

            # write the labels
            label_key = '%s_label%d' % (key, i)
            self._write_array(group, label_key, lab)

    def _read_multi_index(self, group, key):
        nlevels = getattr(group._v_attrs, '%s_nlevels' % key)

        levels = []
        labels = []
        names = []
        for i in range(nlevels):
            level_key = '%s_level%d' % (key, i)
            name, lev = self._read_index_node(getattr(group, level_key))
            levels.append(lev)
            names.append(name)

            label_key = '%s_label%d' % (key, i)
            lab = _read_array(group, label_key)
            labels.append(lab)

        return MultiIndex(levels=levels, labels=labels, names=names)

    def _read_index_node(self, node):
        data = node[:]
        kind = node._v_attrs.kind
        name = None

        if 'name' in node._v_attrs:
            name = node._v_attrs.name

        index_class = _alias_to_class(getattr(node._v_attrs,
                                              'index_class', ''))
        factory = _get_index_factory(index_class)

        kwargs = {}
        if 'freq' in node._v_attrs:
            kwargs['freq'] = node._v_attrs['freq']

        if 'tz' in node._v_attrs:
            kwargs['tz'] = node._v_attrs['tz']

        if kind in ('date', 'datetime'):
            index = factory(_unconvert_index(data, kind), dtype=object,
                            **kwargs)
        else:
            index = factory(_unconvert_index(data, kind), **kwargs)

        index.name = name

        return name, index

    def _write_array(self, group, key, value):
        if key in group:
            self.handle.removeNode(group, key)

        # Transform needed to interface with pytables row/col notation
        empty_array = any(x == 0 for x in value.shape)
        transposed = False
        if not empty_array:
            value = value.T
            transposed = True

        if self.filters is not None:
            atom = None
            try:
                # get the atom for this datatype
                atom = _tables().Atom.from_dtype(value.dtype)
            except ValueError:
                pass

            if atom is not None:
                # create an empty chunked array and fill it from value
                ca = self.handle.createCArray(group, key, atom,
                                              value.shape,
                                              filters=self.filters)
                ca[:] = value
                getattr(group, key)._v_attrs.transposed = transposed
                return

        if value.dtype.type == np.object_:
            vlarr = self.handle.createVLArray(group, key,
                                              _tables().ObjectAtom())
            vlarr.append(value)
        elif value.dtype.type == np.datetime64:
            self.handle.createArray(group, key, value.view('i8'))
            getattr(group, key)._v_attrs.value_type = 'datetime64'
        else:
            if empty_array:
                # ugly hack for length 0 axes
                arr = np.empty((1,) * value.ndim)
                self.handle.createArray(group, key, arr)
                getattr(group, key)._v_attrs.value_type = str(value.dtype)
                getattr(group, key)._v_attrs.shape = value.shape
            else:
                self.handle.createArray(group, key, value)

        getattr(group, key)._v_attrs.transposed = transposed

    def _write_table(self, group, items=None, index=None, columns=None,
                     values=None, append=False, compression=None):
        """ need to check for conform to the existing table:
            e.g. columns should match """
        # create dict of types
        index_converted, index_kind, index_t = _convert_index(index)
        columns_converted, cols_kind, col_t = _convert_index(columns)

        # create the table if it doesn't exist (or get it if it does)
        if not append:
            if 'table' in group:
                self.handle.removeNode(group, 'table')

        if 'table' not in group:
            # create the table
            desc = {'index': index_t,
                    'column': col_t,
                    'values': _tables().FloatCol(shape=(len(values)))}

            options = {'name': 'table',
                       'description': desc}

            if compression:
                complevel = self.complevel
                if complevel is None:
                    complevel = 9
                filters = _tables().Filters(complevel=complevel,
                                            complib=compression,
                                            fletcher32=self.fletcher32)
                options['filters'] = filters
            elif self.filters is not None:
                options['filters'] = self.filters

            table = self.handle.createTable(group, **options)
        else:
            # the table must already exist
            table = getattr(group, 'table', None)

        # check for backwards incompatibility
        if append:
            existing_kind = table._v_attrs.index_kind
            if existing_kind != index_kind:
                raise TypeError("incompatible kind in index [%s - %s]" %
                                (existing_kind, index_kind))

        # add kinds
        table._v_attrs.index_kind = index_kind
        table._v_attrs.columns_kind = cols_kind
        if append:
            existing_fields = getattr(table._v_attrs, 'fields', None)
            if (existing_fields is not None and
                existing_fields != list(items)):
                raise Exception("appended items do not match existing items"
                                " in table!")
        # this depends on creation order of the table
        table._v_attrs.fields = list(items)

        # add the rows
        try:
            for i, index in enumerate(index_converted):
                for c, col in enumerate(columns_converted):
                    v = values[:, i, c]

                    # don't store the row if all values are np.nan
                    if np.isnan(v).all():
                        continue

                    row = table.row
                    row['index'] = index
                    row['column'] = col

                    # create the values array
                    row['values'] = v
                    row.append()
            self.handle.flush()
        except (ValueError), detail:  # pragma: no cover
            print "value_error in _write_table -> %s" % str(detail)
            try:
                self.handle.flush()
            except Exception:
                pass
            raise

    def _read_group(self, group, where=None):
        kind = group._v_attrs.pandas_type
        kind = _LEGACY_MAP.get(kind, kind)
        handler = self._get_handler(op='read', kind=kind)
        return handler(group, where)

    def _read_series(self, group, where=None):
        index = self._read_index(group, 'index')
        if len(index) > 0:
            values = _read_array(group, 'values')
        else:
            values = []

        name = getattr(group._v_attrs, 'name', None)
        return Series(values, index=index, name=name)

    def _read_legacy_series(self, group, where=None):
        index = self._read_index_legacy(group, 'index')
        values = _read_array(group, 'values')
        return Series(values, index=index)

    def _read_legacy_frame(self, group, where=None):
        index = self._read_index_legacy(group, 'index')
        columns = self._read_index_legacy(group, 'columns')
        values = _read_array(group, 'values')
        return DataFrame(values, index=index, columns=columns)

    def _read_index_legacy(self, group, key):
        node = getattr(group, key)
        data = node[:]
        kind = node._v_attrs.kind

        return _unconvert_index_legacy(data, kind)

    def _read_frame_table(self, group, where=None):
        return self._read_panel_table(group, where)['value']

    def _read_panel_table(self, group, where=None):
        table = getattr(group, 'table')
        fields = table._v_attrs.fields

        # create the selection
        sel = Selection(table, where, table._v_attrs.index_kind)
        sel.select()
        fields = table._v_attrs.fields

        columns = _maybe_convert(sel.values['column'],
                                 table._v_attrs.columns_kind)
        index = _maybe_convert(sel.values['index'], table._v_attrs.index_kind)
        values = sel.values['values']

        major = Factor.from_array(index)
        minor = Factor.from_array(columns)

        J, K = len(major.levels), len(minor.levels)
        key = major.labels * K + minor.labels

        if len(unique(key)) == len(key):
            sorter, _ = lib.groupsort_indexer(com._ensure_int64(key), J * K)
            sorter = com._ensure_platform_int(sorter)

            # the data need to be sorted
            sorted_values = values.take(sorter, axis=0)
            major_labels = major.labels.take(sorter)
            minor_labels = minor.labels.take(sorter)

            block = block2d_to_block3d(sorted_values, fields, (J, K),
                                       major_labels, minor_labels)

            mgr = BlockManager([block], [block.ref_items,
                                         major.levels, minor.levels])
            wp = Panel(mgr)
        else:
            if not self._quiet:  # pragma: no cover
                print ('Duplicate entries in table, taking most recently '
                       'appended')

            # reconstruct
            long_index = MultiIndex.from_arrays([index, columns])
            lp = DataFrame(values, index=long_index, columns=fields)

            # need a better algorithm
            tuple_index = long_index._tuple_index

            unique_tuples = lib.fast_unique(tuple_index)
            unique_tuples = _asarray_tuplesafe(unique_tuples)

            indexer = match(unique_tuples, tuple_index)
            indexer = com._ensure_platform_int(indexer)

            new_index = long_index.take(indexer)
            new_values = lp.values.take(indexer, axis=0)

            lp = DataFrame(new_values, index=new_index, columns=lp.columns)
            wp = lp.to_panel()

        if sel.filter:
            new_minor = sorted(set(wp.minor_axis) & sel.filter)
            wp = wp.reindex(minor=new_minor)
        return wp


    def _delete_from_table(self, group, where):
        """ delete rows from a group where condition is True """
        table = getattr(group, 'table')

        # create the selection
        s = Selection(table,where,table._v_attrs.index_kind)
        s.select_coords()

        # delete the rows in reverse order
        l  = list(s.values)
        ln = len(l)

        if ln:

            # if we can do a consecutive removal - do it!
            if l[0]+ln-1 == l[-1]:
                table.removeRows(start = l[0], stop = l[-1]+1)

            # one by one
            else:
                l.reverse()
                for c in l:
                    table.removeRows(c)
                    
                    self.handle.flush()

        # return the number of rows removed
        return ln

def _convert_index(index):
    if isinstance(index, DatetimeIndex):
        converted = index.asi8
        return converted, 'datetime64', _tables().Int64Col()
    elif isinstance(index, (Int64Index, PeriodIndex)):
        atom = _tables().Int64Col()
        return index.values, 'integer', atom

    if isinstance(index, MultiIndex):
        raise Exception('MultiIndex not supported here!')

    inferred_type = lib.infer_dtype(index)

    values = np.asarray(index)

    if inferred_type == 'datetime64':
        converted = values.view('i8')
        return converted, 'datetime64', _tables().Int64Col()
    elif inferred_type == 'datetime':
        converted = np.array([(time.mktime(v.timetuple()) +
                            v.microsecond / 1E6) for v in values],
                            dtype=np.float64)
        return converted, 'datetime', _tables().Time64Col()
    elif inferred_type == 'date':
        converted = np.array([time.mktime(v.timetuple()) for v in values],
                            dtype=np.int32)
        return converted, 'date', _tables().Time32Col()
    elif inferred_type == 'string':
        converted = np.array(list(values), dtype=np.str_)
        itemsize = converted.dtype.itemsize
        return converted, 'string', _tables().StringCol(itemsize)
    elif inferred_type == 'unicode':
        atom = _tables().ObjectAtom()
        return np.asarray(values, dtype='O'), 'object', atom
    elif inferred_type == 'integer':
        # take a guess for now, hope the values fit
        atom = _tables().Int64Col()
        return np.asarray(values, dtype=np.int64), 'integer', atom
    elif inferred_type == 'floating':
        atom = _tables().Float64Col()
        return np.asarray(values, dtype=np.float64), 'float', atom
    else:  # pragma: no cover
        atom = _tables().ObjectAtom()
        return np.asarray(values, dtype='O'), 'object', atom


def _read_array(group, key):
    import tables
    node = getattr(group, key)
    data = node[:]
    attrs = node._v_attrs

    transposed = getattr(attrs, 'transposed', False)

    if isinstance(node, tables.VLArray):
        ret = data[0]
    else:
        dtype = getattr(attrs, 'value_type', None)
        shape = getattr(attrs, 'shape', None)

        if shape is not None:
            # length 0 axis
            ret = np.empty(shape, dtype=dtype)
        else:
            ret = data

        if dtype == 'datetime64':
            ret = np.array(ret, dtype='M8[ns]')

    if transposed:
        return ret.T
    else:
        return ret


def _unconvert_index(data, kind):
    if kind == 'datetime64':
        index = DatetimeIndex(data)
    elif kind == 'datetime':
        index = np.array([datetime.fromtimestamp(v) for v in data],
                         dtype=object)
    elif kind == 'date':
        index = np.array([date.fromtimestamp(v) for v in data], dtype=object)
    elif kind in ('string', 'integer', 'float'):
        index = np.array(data)
    elif kind == 'object':
        index = np.array(data[0])
    else:  # pragma: no cover
        raise ValueError('unrecognized index type %s' % kind)
    return index


def _unconvert_index_legacy(data, kind, legacy=False):
    if kind == 'datetime':
        index = lib.time64_to_datetime(data)
    elif kind in ('string', 'integer'):
        index = np.array(data, dtype=object)
    else:  # pragma: no cover
        raise ValueError('unrecognized index type %s' % kind)
    return index


def _maybe_convert(values, val_kind):
    if _need_convert(val_kind):
        conv = _get_converter(val_kind)
        # conv = np.frompyfunc(conv, 1, 1)
        values = conv(values)
    return values


def _get_converter(kind):
    if kind == 'datetime64':
        return lambda x: np.array(x, dtype='M8[ns]')
    if kind == 'datetime':
        return lib.convert_timestamps
    else:  # pragma: no cover
        raise ValueError('invalid kind %s' % kind)


def _need_convert(kind):
    if kind in ('datetime', 'datetime64'):
        return True
    return False


def _is_table_type(group):
    try:
        return 'table' in group._v_attrs.pandas_type
    except AttributeError:
        # new node, e.g.
        return False

_index_type_map = {DatetimeIndex: 'datetime',
                   PeriodIndex: 'period'}

_reverse_index_map = {}
for k, v in _index_type_map.iteritems():
    _reverse_index_map[v] = k


def _class_to_alias(cls):
    return _index_type_map.get(cls, '')


def _alias_to_class(alias):
    if isinstance(alias, type):  # pragma: no cover
        return alias  # compat: for a short period of time master stored types
    return _reverse_index_map.get(alias, Index)


class Term(object):
    """ create a term object that holds a field, op, and value 

        Parameters
        ----------
        field : dict, string term expression, or the field to operate (must be a valid index/column type of DataFrame/Panel)
        op    : a valid op (defaults to '=') (optional)
                >, >=, <, <=, =, != (not equal) are allowed
        value : a value or list of values (required)

        Returns
        -------
        a Term object

        Examples
        --------
        Term(dict(field = 'index', op = '>', value = '20121114'))
        Term('index', '20121114')
        Term('index', '>', '20121114')
        Term('index', ['20121114','20121114'])
        Term('index', datetime(2012,11,14))
        Term('index>20121114')
        
    """

    _ops     = ['<','<=','>','>=','=','!=']
    _search  = re.compile("^(?P<field>\w+)(?P<op>%s)(?P<value>.+)$" % '|'.join(_ops))
    _index   = ['index','major_axis']
    _column  = ['column','minor_axis','items']

    def __init__(self, field, op = None, value = None, index_kind = None):
        self.field = None
        self.op    = None
        self.value = None
        self.typ   = None
        self.index_kind = index_kind
        self.filter     = None
        self.condition  = None

        # unpack lists/tuples in field
        if isinstance(field,(tuple,list)):
            f = field
            field = f[0]
            if len(f) > 1:
                op = f[1]
            if len(f) > 2:
                value = f[2]
                
        # backwards compatible
        if isinstance(field, dict):
            self.field = field.get('field')
            self.op    = field.get('op') or '='
            self.value = field.get('value')

        # passed a term
        elif isinstance(field,Term):
            self.field = field.field
            self.op    = field.op
            self.value = field.value

        # a string expression (or just the field)
        elif isinstance(field,basestring):

            # is a term is passed
            s = self._search.match(field)
            if s is not None:
                self.field = s.group('field')
                self.op    = s.group('op')
                self.value = s.group('value')

            else:
                self.field = field

                # is an op passed?
                if isinstance(op, basestring) and op in self._ops:
                    self.op    = op
                    self.value = value
                else:
                    self.op    = '='
                    self.value = op

        else:
            raise Exception("Term does not understand the supplied field [%s]" % field)
        
        # we have valid fields
        if self.field is None or self.op is None or self.value is None:
            raise Exception("Could not create this term [%s]" % str(self))

        # valid field name
        if self.field in self._index:
            self.typ = 'index'
        elif self.field in self._column:
            self.typ = 'column'
        else:
            raise Exception("field is not a valid index/column for this term [%s]" % str(self))

        # we have valid conditions
        if self.op in ['>','>=','<','<=']:
            if hasattr(self.value,'__iter__') and len(self.value) > 1:
                raise Exception("an inequality condition cannot have multiple values [%s]" % str(self))

        if not hasattr(self.value,'__iter__'):
            self.value = [ self.value ]

        self.eval()

    def __str__(self):
        return "typ->%s,field->%s,op->%s,value->%s" % (self.typ,self.field,self.op,self.value)

    __repr__ = __str__

    def eval(self):
        """ set the numexpr expression for this term """
        
        # convert values
        values = [ self.convert_value(v) for v in self.value ]

        # equality conditions
        if self.op in ['=','!=']:
        
            # too many values to create the expression?
            if len(values) <= 61:
                self.condition = "(%s)" % ' | '.join([ "(%s == %s)" % (self.field,v[0]) for v in values])

            # use a filter after reading
            else:
                self.filter = set([ v[1] for v in values ])

        else:

            self.condition = '(%s %s %s)' % (self.field, self.op, values[0][0])
      
    def convert_value(self, v):

        if self.typ == 'index':
            if self.index_kind == 'datetime64' :
                return [lib.Timestamp(v).value, None]
            elif isinstance(v, datetime):
                return [time.mktime(v.timetuple()), None]
        elif not isinstance(v, basestring):
            return [str(v), None]

        # string quoting
        return ["'" + v + "'", v]

class Selection(object):
    """
    Carries out a selection operation on a tables.Table object.

    Parameters
    ----------
    table : tables.Table
    where : list of Terms (or convertable to)

    """
    def __init__(self, table, where=None, index_kind=None):
        self.table      = table
        self.where      = where
        self.index_kind = index_kind
        self.values     = None
        self.condition  = None
        self.filter     = None
        self.terms      = self.generate(where)

        # create the numexpr & the filter
        if self.terms:
            conds = [ t.condition for t in self.terms if t.condition is not None ]
            if len(conds):
                self.condition = "(%s)" % ' & '.join(conds)
            self.filter    = set()
            for t in self.terms:
                if t.filter is not None:
                    self.filter |= t.filter

    def generate(self, where):
        """ generate and return the terms """
        if where is None: return None

        if not isinstance(where, (list,tuple)):
            where = [ where ]

        return [ Term(c, index_kind = self.index_kind) for c in where ]

    def select(self):
        """
        generate the selection
        """
        if self.condition is not None:
            self.values = self.table.readWhere(self.condition)
        else:
            self.values = self.table.read()

    def select_coords(self):
        """
        generate the selection
        """
        self.values = self.table.getWhereList(self.condition)


def _get_index_factory(klass):
    if klass == DatetimeIndex:
        def f(values, freq=None, tz=None):
            return DatetimeIndex._simple_new(values, None, freq=freq,
                                             tz=tz)
        return f
    return klass