ops.py

"""
Arithmetic operations for PandasObjects

This is not a public API.
"""
# necessary to enforce truediv in Python 2.X
from __future__ import division
import operator
import warnings
import numpy as np
import pandas as pd
import datetime

from pandas._libs import (lib, index as libindex,
                          tslib as libts, algos as libalgos, iNaT)

from pandas import compat
from pandas.util.decorators import Appender
import pandas.computation.expressions as expressions

from pandas.compat import bind_method
import pandas.core.missing as missing

from pandas.errors import PerformanceWarning
from pandas.core.common import _values_from_object, _maybe_match_name
from pandas.types.missing import notnull, isnull
from pandas.types.common import (needs_i8_conversion,
                                 is_datetimelike_v_numeric,
                                 is_integer_dtype, is_categorical_dtype,
                                 is_object_dtype, is_timedelta64_dtype,
                                 is_datetime64_dtype, is_datetime64tz_dtype,
                                 is_bool_dtype, is_datetimetz,
                                 is_list_like,
                                 is_scalar,
                                 _ensure_object)
from pandas.types.cast import maybe_upcast_putmask, find_common_type
from pandas.types.generic import ABCSeries, ABCIndex, ABCPeriodIndex

# -----------------------------------------------------------------------------
# Functions that add arithmetic methods to objects, given arithmetic factory
# methods


def _create_methods(arith_method, comp_method, bool_method,
                    use_numexpr, special=False, default_axis='columns',
                    have_divmod=False):
    # creates actual methods based upon arithmetic, comp and bool method
    # constructors.

    # NOTE: Only frame cares about default_axis, specifically: special methods
    # have default axis None, whereas flex methods have default axis 'columns'
    # if we're not using numexpr, then don't pass a str_rep
    if use_numexpr:
        op = lambda x: x
    else:
        op = lambda x: None
    if special:

        def names(x):
            if x[-1] == "_":
                return "__%s_" % x
            else:
                return "__%s__" % x
    else:
        names = lambda x: x

    # Inframe, all special methods have default_axis=None, flex methods have
    # default_axis set to the default (columns)
    # yapf: disable
    new_methods = dict(
        add=arith_method(operator.add, names('add'), op('+'),
                         default_axis=default_axis),
        radd=arith_method(lambda x, y: y + x, names('radd'), op('+'),
                          default_axis=default_axis),
        sub=arith_method(operator.sub, names('sub'), op('-'),
                         default_axis=default_axis),
        mul=arith_method(operator.mul, names('mul'), op('*'),
                         default_axis=default_axis),
        truediv=arith_method(operator.truediv, names('truediv'), op('/'),
                             truediv=True, fill_zeros=np.inf,
                             default_axis=default_axis),
        floordiv=arith_method(operator.floordiv, names('floordiv'), op('//'),
                              default_axis=default_axis, fill_zeros=np.inf),
        # Causes a floating point exception in the tests when numexpr enabled,
        # so for now no speedup
        mod=arith_method(operator.mod, names('mod'), None,
                         default_axis=default_axis, fill_zeros=np.nan),
        pow=arith_method(operator.pow, names('pow'), op('**'),
                         default_axis=default_axis),
        # not entirely sure why this is necessary, but previously was included
        # so it's here to maintain compatibility
        rmul=arith_method(operator.mul, names('rmul'), op('*'),
                          default_axis=default_axis, reversed=True),
        rsub=arith_method(lambda x, y: y - x, names('rsub'), op('-'),
                          default_axis=default_axis, reversed=True),
        rtruediv=arith_method(lambda x, y: operator.truediv(y, x),
                              names('rtruediv'), op('/'), truediv=True,
                              fill_zeros=np.inf, default_axis=default_axis,
                              reversed=True),
        rfloordiv=arith_method(lambda x, y: operator.floordiv(y, x),
                               names('rfloordiv'), op('//'),
                               default_axis=default_axis, fill_zeros=np.inf,
                               reversed=True),
        rpow=arith_method(lambda x, y: y**x, names('rpow'), op('**'),
                          default_axis=default_axis, reversed=True),
        rmod=arith_method(lambda x, y: y % x, names('rmod'), op('%'),
                          default_axis=default_axis, fill_zeros=np.nan,
                          reversed=True),)
    # yapf: enable
    new_methods['div'] = new_methods['truediv']
    new_methods['rdiv'] = new_methods['rtruediv']

    # Comp methods never had a default axis set
    if comp_method:
        new_methods.update(dict(
            eq=comp_method(operator.eq, names('eq'), op('==')),
            ne=comp_method(operator.ne, names('ne'), op('!='), masker=True),
            lt=comp_method(operator.lt, names('lt'), op('<')),
            gt=comp_method(operator.gt, names('gt'), op('>')),
            le=comp_method(operator.le, names('le'), op('<=')),
            ge=comp_method(operator.ge, names('ge'), op('>=')), ))
    if bool_method:
        new_methods.update(
            dict(and_=bool_method(operator.and_, names('and_'), op('&')),
                 or_=bool_method(operator.or_, names('or_'), op('|')),
                 # For some reason ``^`` wasn't used in original.
                 xor=bool_method(operator.xor, names('xor'), op('^')),
                 rand_=bool_method(lambda x, y: operator.and_(y, x),
                                   names('rand_'), op('&')),
                 ror_=bool_method(lambda x, y: operator.or_(y, x),
                                  names('ror_'), op('|')),
                 rxor=bool_method(lambda x, y: operator.xor(y, x),
                                  names('rxor'), op('^'))))
    if have_divmod:
        # divmod doesn't have an op that is supported by numexpr
        new_methods['divmod'] = arith_method(
            divmod,
            names('divmod'),
            None,
            default_axis=default_axis,
            construct_result=_construct_divmod_result,
        )

    new_methods = dict((names(k), v) for k, v in new_methods.items())
    return new_methods


def add_methods(cls, new_methods, force, select, exclude):
    if select and exclude:
        raise TypeError("May only pass either select or exclude")
    methods = new_methods
    if select:
        select = set(select)
        methods = {}
        for key, method in new_methods.items():
            if key in select:
                methods[key] = method
    if exclude:
        for k in exclude:
            new_methods.pop(k, None)

    for name, method in new_methods.items():
        if force or name not in cls.__dict__:
            bind_method(cls, name, method)


# ----------------------------------------------------------------------
# Arithmetic
def add_special_arithmetic_methods(cls, arith_method=None,
                                   comp_method=None, bool_method=None,
                                   use_numexpr=True, force=False, select=None,
                                   exclude=None, have_divmod=False):
    """
    Adds the full suite of special arithmetic methods (``__add__``,
    ``__sub__``, etc.) to the class.

    Parameters
    ----------
    arith_method : function (optional)
        factory for special arithmetic methods, with op string:
        f(op, name, str_rep, default_axis=None, fill_zeros=None, **eval_kwargs)
    comp_method : function, optional,
        factory for rich comparison - signature: f(op, name, str_rep)
    use_numexpr : bool, default True
        whether to accelerate with numexpr, defaults to True
    force : bool, default False
        if False, checks whether function is defined **on ``cls.__dict__``**
        before defining if True, always defines functions on class base
    select : iterable of strings (optional)
        if passed, only sets functions with names in select
    exclude : iterable of strings (optional)
        if passed, will not set functions with names in exclude
    have_divmod : bool, (optional)
        should a divmod method be added? this method is special because it
        returns a tuple of cls instead of a single element of type cls
    """

    # in frame, special methods have default_axis = None, comp methods use
    # 'columns'

    new_methods = _create_methods(arith_method, comp_method,
                                  bool_method, use_numexpr, default_axis=None,
                                  special=True, have_divmod=have_divmod)

    # inplace operators (I feel like these should get passed an `inplace=True`
    # or just be removed

    def _wrap_inplace_method(method):
        """
        return an inplace wrapper for this method
        """

        def f(self, other):
            result = method(self, other)

            # this makes sure that we are aligned like the input
            # we are updating inplace so we want to ignore is_copy
            self._update_inplace(result.reindex_like(self, copy=False)._data,
                                 verify_is_copy=False)

            return self

        return f

    new_methods.update(
        dict(__iadd__=_wrap_inplace_method(new_methods["__add__"]),
             __isub__=_wrap_inplace_method(new_methods["__sub__"]),
             __imul__=_wrap_inplace_method(new_methods["__mul__"]),
             __itruediv__=_wrap_inplace_method(new_methods["__truediv__"]),
             __ipow__=_wrap_inplace_method(new_methods["__pow__"]), ))
    if not compat.PY3:
        new_methods["__idiv__"] = new_methods["__div__"]

    add_methods(cls, new_methods=new_methods, force=force, select=select,
                exclude=exclude)


def add_flex_arithmetic_methods(cls, flex_arith_method,
                                flex_comp_method=None, flex_bool_method=None,
                                use_numexpr=True, force=False, select=None,
                                exclude=None):
    """
    Adds the full suite of flex arithmetic methods (``pow``, ``mul``, ``add``)
    to the class.

    Parameters
    ----------
    flex_arith_method : function
        factory for special arithmetic methods, with op string:
        f(op, name, str_rep, default_axis=None, fill_zeros=None, **eval_kwargs)
    flex_comp_method : function, optional,
        factory for rich comparison - signature: f(op, name, str_rep)
    use_numexpr : bool, default True
        whether to accelerate with numexpr, defaults to True
    force : bool, default False
        if False, checks whether function is defined **on ``cls.__dict__``**
        before defining if True, always defines functions on class base
    select : iterable of strings (optional)
        if passed, only sets functions with names in select
    exclude : iterable of strings (optional)
        if passed, will not set functions with names in exclude
    """
    # in frame, default axis is 'columns', doesn't matter for series and panel
    new_methods = _create_methods(flex_arith_method,
                                  flex_comp_method, flex_bool_method,
                                  use_numexpr, default_axis='columns',
                                  special=False)
    new_methods.update(dict(multiply=new_methods['mul'],
                            subtract=new_methods['sub'],
                            divide=new_methods['div']))
    # opt out of bool flex methods for now
    for k in ('ror_', 'rxor', 'rand_'):
        if k in new_methods:
            new_methods.pop(k)

    add_methods(cls, new_methods=new_methods, force=force, select=select,
                exclude=exclude)


class _Op(object):

    """
    Wrapper around Series arithmetic operations.
    Generally, you should use classmethod ``_Op.get_op`` as an entry point.

    This validates and coerces lhs and rhs depending on its dtype and
    based on op. See _TimeOp also.

    Parameters
    ----------
    left : Series
        lhs of op
    right : object
        rhs of op
    name : str
        name of op
    na_op : callable
        a function which wraps op
    """

    fill_value = np.nan
    wrap_results = staticmethod(lambda x: x)
    dtype = None

    def __init__(self, left, right, name, na_op):
        self.left = left
        self.right = right

        self.name = name
        self.na_op = na_op

        self.lvalues = left
        self.rvalues = right

    @classmethod
    def get_op(cls, left, right, name, na_op):
        """
        Get op dispatcher, returns _Op or _TimeOp.

        If ``left`` and ``right`` are appropriate for datetime arithmetic with
        operation ``name``, processes them and returns a ``_TimeOp`` object
        that stores all the required values.  Otherwise, it will generate
        either a ``_Op``, indicating that the operation is performed via
        normal numpy path.
        """
        is_timedelta_lhs = is_timedelta64_dtype(left)
        is_datetime_lhs = (is_datetime64_dtype(left) or
                           is_datetime64tz_dtype(left))

        if not (is_datetime_lhs or is_timedelta_lhs):
            return _Op(left, right, name, na_op)
        else:
            return _TimeOp(left, right, name, na_op)


class _TimeOp(_Op):
    """
    Wrapper around Series datetime/time/timedelta arithmetic operations.
    Generally, you should use classmethod ``_Op.get_op`` as an entry point.
    """
    fill_value = iNaT

    def __init__(self, left, right, name, na_op):
        super(_TimeOp, self).__init__(left, right, name, na_op)

        lvalues = self._convert_to_array(left, name=name)
        rvalues = self._convert_to_array(right, name=name, other=lvalues)

        # left
        self.is_offset_lhs = self._is_offset(left)
        self.is_timedelta_lhs = is_timedelta64_dtype(lvalues)
        self.is_datetime64_lhs = is_datetime64_dtype(lvalues)
        self.is_datetime64tz_lhs = is_datetime64tz_dtype(lvalues)
        self.is_datetime_lhs = (self.is_datetime64_lhs or
                                self.is_datetime64tz_lhs)
        self.is_integer_lhs = left.dtype.kind in ['i', 'u']
        self.is_floating_lhs = left.dtype.kind == 'f'

        # right
        self.is_offset_rhs = self._is_offset(right)
        self.is_datetime64_rhs = is_datetime64_dtype(rvalues)
        self.is_datetime64tz_rhs = is_datetime64tz_dtype(rvalues)
        self.is_datetime_rhs = (self.is_datetime64_rhs or
                                self.is_datetime64tz_rhs)
        self.is_timedelta_rhs = is_timedelta64_dtype(rvalues)
        self.is_integer_rhs = rvalues.dtype.kind in ('i', 'u')
        self.is_floating_rhs = rvalues.dtype.kind == 'f'

        self._validate(lvalues, rvalues, name)
        self.lvalues, self.rvalues = self._convert_for_datetime(lvalues,
                                                                rvalues)

    def _validate(self, lvalues, rvalues, name):
        # timedelta and integer mul/div

        if ((self.is_timedelta_lhs and
                (self.is_integer_rhs or self.is_floating_rhs)) or
            (self.is_timedelta_rhs and
                (self.is_integer_lhs or self.is_floating_lhs))):

            if name not in ('__div__', '__truediv__', '__mul__', '__rmul__'):
                raise TypeError("can only operate on a timedelta and an "
                                "integer or a float for division and "
                                "multiplication, but the operator [%s] was"
                                "passed" % name)

        # 2 timedeltas
        elif ((self.is_timedelta_lhs and
               (self.is_timedelta_rhs or self.is_offset_rhs)) or
              (self.is_timedelta_rhs and
               (self.is_timedelta_lhs or self.is_offset_lhs))):

            if name not in ('__div__', '__rdiv__', '__truediv__',
                            '__rtruediv__', '__add__', '__radd__', '__sub__',
                            '__rsub__'):
                raise TypeError("can only operate on a timedeltas for "
                                "addition, subtraction, and division, but the"
                                " operator [%s] was passed" % name)

        # datetime and timedelta/DateOffset
        elif (self.is_datetime_lhs and
              (self.is_timedelta_rhs or self.is_offset_rhs)):

            if name not in ('__add__', '__radd__', '__sub__'):
                raise TypeError("can only operate on a datetime with a rhs of "
                                "a timedelta/DateOffset for addition and "
                                "subtraction, but the operator [%s] was "
                                "passed" % name)

        elif (self.is_datetime_rhs and
              (self.is_timedelta_lhs or self.is_offset_lhs)):
            if name not in ('__add__', '__radd__', '__rsub__'):
                raise TypeError("can only operate on a timedelta/DateOffset "
                                "with a rhs of a datetime for addition, "
                                "but the operator [%s] was passed" % name)

        # 2 datetimes
        elif self.is_datetime_lhs and self.is_datetime_rhs:

            if name not in ('__sub__', '__rsub__'):
                raise TypeError("can only operate on a datetimes for"
                                " subtraction, but the operator [%s] was"
                                " passed" % name)

            # if tz's must be equal (same or None)
            if getattr(lvalues, 'tz', None) != getattr(rvalues, 'tz', None):
                raise ValueError("Incompatible tz's on datetime subtraction "
                                 "ops")

        elif ((self.is_timedelta_lhs or self.is_offset_lhs) and
              self.is_datetime_rhs):

            if name not in ('__add__', '__radd__'):
                raise TypeError("can only operate on a timedelta/DateOffset "
                                "and a datetime for addition, but the "
                                "operator [%s] was passed" % name)
        else:
            raise TypeError('cannot operate on a series without a rhs '
                            'of a series/ndarray of type datetime64[ns] '
                            'or a timedelta')

    def _convert_to_array(self, values, name=None, other=None):
        """converts values to ndarray"""
        from pandas.tseries.timedeltas import to_timedelta

        ovalues = values
        supplied_dtype = None
        if not is_list_like(values):
            values = np.array([values])

        # if this is a Series that contains relevant dtype info, then use this
        # instead of the inferred type; this avoids coercing Series([NaT],
        # dtype='datetime64[ns]') to Series([NaT], dtype='timedelta64[ns]')
        elif (isinstance(values, pd.Series) and
              (is_timedelta64_dtype(values) or is_datetime64_dtype(values))):
            supplied_dtype = values.dtype

        inferred_type = lib.infer_dtype(values)
        if (inferred_type in ('datetime64', 'datetime', 'date', 'time') or
                is_datetimetz(inferred_type)):
            # if we have a other of timedelta, but use pd.NaT here we
            # we are in the wrong path
            if (supplied_dtype is None and other is not None and
                (other.dtype in ('timedelta64[ns]', 'datetime64[ns]')) and
                    isnull(values).all()):
                values = np.empty(values.shape, dtype='timedelta64[ns]')
                values[:] = iNaT

            # a datelike
            elif isinstance(values, pd.DatetimeIndex):
                values = values.to_series()
            # datetime with tz
            elif (isinstance(ovalues, datetime.datetime) and
                  hasattr(ovalues, 'tzinfo')):
                values = pd.DatetimeIndex(values)
            # datetime array with tz
            elif is_datetimetz(values):
                if isinstance(values, ABCSeries):
                    values = values._values
            elif not (isinstance(values, (np.ndarray, ABCSeries)) and
                      is_datetime64_dtype(values)):
                values = libts.array_to_datetime(values)
        elif inferred_type in ('timedelta', 'timedelta64'):
            # have a timedelta, convert to to ns here
            values = to_timedelta(values, errors='coerce', box=False)
        elif inferred_type == 'integer':
            # py3 compat where dtype is 'm' but is an integer
            if values.dtype.kind == 'm':
                values = values.astype('timedelta64[ns]')
            elif isinstance(values, pd.PeriodIndex):
                values = values.to_timestamp().to_series()
            elif name not in ('__truediv__', '__div__', '__mul__', '__rmul__'):
                raise TypeError("incompatible type for a datetime/timedelta "
                                "operation [{0}]".format(name))
        elif inferred_type == 'floating':
            if (isnull(values).all() and
                    name in ('__add__', '__radd__', '__sub__', '__rsub__')):
                values = np.empty(values.shape, dtype=other.dtype)
                values[:] = iNaT
            return values
        elif self._is_offset(values):
            return values
        else:
            raise TypeError("incompatible type [{0}] for a datetime/timedelta"
                            " operation".format(np.array(values).dtype))

        return values

    def _convert_for_datetime(self, lvalues, rvalues):
        from pandas.tseries.timedeltas import to_timedelta

        mask = isnull(lvalues) | isnull(rvalues)

        # datetimes require views
        if self.is_datetime_lhs or self.is_datetime_rhs:

            # datetime subtraction means timedelta
            if self.is_datetime_lhs and self.is_datetime_rhs:
                if self.name in ('__sub__', '__rsub__'):
                    self.dtype = 'timedelta64[ns]'
                else:
                    self.dtype = 'datetime64[ns]'
            elif self.is_datetime64tz_lhs:
                self.dtype = lvalues.dtype
            elif self.is_datetime64tz_rhs:
                self.dtype = rvalues.dtype
            else:
                self.dtype = 'datetime64[ns]'

            # if adding single offset try vectorized path
            # in DatetimeIndex; otherwise elementwise apply
            def _offset(lvalues, rvalues):
                if len(lvalues) == 1:
                    rvalues = pd.DatetimeIndex(rvalues)
                    lvalues = lvalues[0]
                else:
                    warnings.warn("Adding/subtracting array of DateOffsets to "
                                  "Series not vectorized", PerformanceWarning)
                    rvalues = rvalues.astype('O')

                # pass thru on the na_op
                self.na_op = lambda x, y: getattr(x, self.name)(y)
                return lvalues, rvalues

            if self.is_offset_lhs:
                lvalues, rvalues = _offset(lvalues, rvalues)
            elif self.is_offset_rhs:
                rvalues, lvalues = _offset(rvalues, lvalues)
            else:

                # with tz, convert to UTC
                if self.is_datetime64tz_lhs:
                    lvalues = lvalues.tz_convert('UTC').tz_localize(None)
                if self.is_datetime64tz_rhs:
                    rvalues = rvalues.tz_convert('UTC').tz_localize(None)

                lvalues = lvalues.view(np.int64)
                rvalues = rvalues.view(np.int64)

        # otherwise it's a timedelta
        else:

            self.dtype = 'timedelta64[ns]'

            # convert Tick DateOffset to underlying delta
            if self.is_offset_lhs:
                lvalues = to_timedelta(lvalues, box=False)
            if self.is_offset_rhs:
                rvalues = to_timedelta(rvalues, box=False)

            lvalues = lvalues.astype(np.int64)
            if not self.is_floating_rhs:
                rvalues = rvalues.astype(np.int64)

            # time delta division -> unit less
            # integer gets converted to timedelta in np < 1.6
            if ((self.is_timedelta_lhs and self.is_timedelta_rhs) and
                    not self.is_integer_rhs and not self.is_integer_lhs and
                    self.name in ('__div__', '__truediv__')):
                self.dtype = 'float64'
                self.fill_value = np.nan
                lvalues = lvalues.astype(np.float64)
                rvalues = rvalues.astype(np.float64)

        # if we need to mask the results
        if mask.any():

            def f(x):

                # datetime64[ns]/timedelta64[ns] masking
                try:
                    x = np.array(x, dtype=self.dtype)
                except TypeError:
                    x = np.array(x, dtype='datetime64[ns]')

                np.putmask(x, mask, self.fill_value)
                return x

            self.wrap_results = f

        return lvalues, rvalues

    def _is_offset(self, arr_or_obj):
        """ check if obj or all elements of list-like is DateOffset """
        if isinstance(arr_or_obj, pd.DateOffset):
            return True
        elif is_list_like(arr_or_obj) and len(arr_or_obj):
            return all(isinstance(x, pd.DateOffset) for x in arr_or_obj)
        return False


def _align_method_SERIES(left, right, align_asobject=False):
    """ align lhs and rhs Series """

    # ToDo: Different from _align_method_FRAME, list, tuple and ndarray
    # are not coerced here
    # because Series has inconsistencies described in #13637

    if isinstance(right, ABCSeries):
        # avoid repeated alignment
        if not left.index.equals(right.index):

            if align_asobject:
                # to keep original value's dtype for bool ops
                left = left.astype(object)
                right = right.astype(object)

            left, right = left.align(right, copy=False)

    return left, right


def _construct_result(left, result, index, name, dtype):
    return left._constructor(result, index=index, name=name, dtype=dtype)


def _construct_divmod_result(left, result, index, name, dtype):
    """divmod returns a tuple of like indexed series instead of a single series.
    """
    constructor = left._constructor
    return (
        constructor(result[0], index=index, name=name, dtype=dtype),
        constructor(result[1], index=index, name=name, dtype=dtype),
    )


def _arith_method_SERIES(op, name, str_rep, fill_zeros=None, default_axis=None,
                         construct_result=_construct_result, **eval_kwargs):
    """
    Wrapper function for Series arithmetic operations, to avoid
    code duplication.
    """

    def na_op(x, y):
        try:
            result = expressions.evaluate(op, str_rep, x, y,
                                          raise_on_error=True, **eval_kwargs)
        except TypeError:
            if isinstance(y, (np.ndarray, ABCSeries, pd.Index)):
                dtype = find_common_type([x.dtype, y.dtype])
                result = np.empty(x.size, dtype=dtype)
                mask = notnull(x) & notnull(y)
                result[mask] = op(x[mask], _values_from_object(y[mask]))
            elif isinstance(x, np.ndarray):
                result = np.empty(len(x), dtype=x.dtype)
                mask = notnull(x)
                result[mask] = op(x[mask], y)
            else:
                raise TypeError("{typ} cannot perform the operation "
                                "{op}".format(typ=type(x).__name__,
                                              op=str_rep))

            result, changed = maybe_upcast_putmask(result, ~mask, np.nan)

        result = missing.fill_zeros(result, x, y, name, fill_zeros)
        return result

    def safe_na_op(lvalues, rvalues):
        try:
            with np.errstate(all='ignore'):
                return na_op(lvalues, rvalues)
        except Exception:
            if isinstance(rvalues, ABCSeries):
                if is_object_dtype(rvalues):
                    # if dtype is object, try elementwise op
                    return libalgos.arrmap_object(rvalues,
                                                  lambda x: op(lvalues, x))
            else:
                if is_object_dtype(lvalues):
                    return libalgos.arrmap_object(lvalues,
                                                  lambda x: op(x, rvalues))
            raise

    def wrapper(left, right, name=name, na_op=na_op):

        if isinstance(right, pd.DataFrame):
            return NotImplemented

        left, right = _align_method_SERIES(left, right)

        converted = _Op.get_op(left, right, name, na_op)

        left, right = converted.left, converted.right
        lvalues, rvalues = converted.lvalues, converted.rvalues
        dtype = converted.dtype
        wrap_results = converted.wrap_results
        na_op = converted.na_op

        if isinstance(rvalues, ABCSeries):
            name = _maybe_match_name(left, rvalues)
            lvalues = getattr(lvalues, 'values', lvalues)
            rvalues = getattr(rvalues, 'values', rvalues)
            # _Op aligns left and right
        else:
            name = left.name
            if (hasattr(lvalues, 'values') and
                    not isinstance(lvalues, pd.DatetimeIndex)):
                lvalues = lvalues.values

        result = wrap_results(safe_na_op(lvalues, rvalues))
        return construct_result(
            left,
            result,
            index=left.index,
            name=name,
            dtype=dtype,
        )

    return wrapper


def _comp_method_OBJECT_ARRAY(op, x, y):
    if isinstance(y, list):
        y = lib.list_to_object_array(y)
    if isinstance(y, (np.ndarray, ABCSeries, ABCIndex)):
        if not is_object_dtype(y.dtype):
            y = y.astype(np.object_)

        if isinstance(y, (ABCSeries, ABCIndex)):
            y = y.values

        result = lib.vec_compare(x, y, op)
    else:
        result = lib.scalar_compare(x, y, op)
    return result


def _comp_method_SERIES(op, name, str_rep, masker=False):
    """
    Wrapper function for Series arithmetic operations, to avoid
    code duplication.
    """

    def na_op(x, y):

        # dispatch to the categorical if we have a categorical
        # in either operand
        if is_categorical_dtype(x):
            return op(x, y)
        elif is_categorical_dtype(y) and not is_scalar(y):
            return op(y, x)

        if is_object_dtype(x.dtype):
            result = _comp_method_OBJECT_ARRAY(op, x, y)
        else:

            # we want to compare like types
            # we only want to convert to integer like if
            # we are not NotImplemented, otherwise
            # we would allow datetime64 (but viewed as i8) against
            # integer comparisons
            if is_datetimelike_v_numeric(x, y):
                raise TypeError("invalid type comparison")

            # numpy does not like comparisons vs None
            if is_scalar(y) and isnull(y):
                if name == '__ne__':
                    return np.ones(len(x), dtype=bool)
                else:
                    return np.zeros(len(x), dtype=bool)

            # we have a datetime/timedelta and may need to convert
            mask = None
            if (needs_i8_conversion(x) or
                    (not is_scalar(y) and needs_i8_conversion(y))):

                if is_scalar(y):
                    mask = isnull(x)
                    y = libindex.convert_scalar(x, _values_from_object(y))
                else:
                    mask = isnull(x) | isnull(y)
                    y = y.view('i8')
                x = x.view('i8')

            try:
                with np.errstate(all='ignore'):
                    result = getattr(x, name)(y)
                if result is NotImplemented:
                    raise TypeError("invalid type comparison")
            except AttributeError:
                result = op(x, y)

            if mask is not None and mask.any():
                result[mask] = masker

        return result

    def wrapper(self, other, axis=None):
        # Validate the axis parameter
        if axis is not None:
            self._get_axis_number(axis)

        if isinstance(other, ABCSeries):
            name = _maybe_match_name(self, other)
            if not self._indexed_same(other):
                msg = 'Can only compare identically-labeled Series objects'
                raise ValueError(msg)
            return self._constructor(na_op(self.values, other.values),
                                     index=self.index, name=name)
        elif isinstance(other, pd.DataFrame):  # pragma: no cover
            return NotImplemented
        elif isinstance(other, (np.ndarray, pd.Index)):
            # do not check length of zerodim array
            # as it will broadcast
            if (not is_scalar(lib.item_from_zerodim(other)) and
                    len(self) != len(other)):
                raise ValueError('Lengths must match to compare')

            if isinstance(other, ABCPeriodIndex):
                # temp workaround until fixing GH 13637
                # tested in test_nat_comparisons
                # (pandas.tests.series.test_operators.TestSeriesOperators)
                return self._constructor(na_op(self.values,
                                               other.asobject.values),
                                         index=self.index)

            return self._constructor(na_op(self.values, np.asarray(other)),
                                     index=self.index).__finalize__(self)

        elif isinstance(other, pd.Categorical):
            if not is_categorical_dtype(self):
                msg = ("Cannot compare a Categorical for op {op} with Series "
                       "of dtype {typ}.\nIf you want to compare values, use "
                       "'series <op> np.asarray(other)'.")
                raise TypeError(msg.format(op=op, typ=self.dtype))

        if is_categorical_dtype(self):
            # cats are a special case as get_values() would return an ndarray,
            # which would then not take categories ordering into account
            # we can go directly to op, as the na_op would just test again and
            # dispatch to it.
            with np.errstate(all='ignore'):
                res = op(self.values, other)
        else:
            values = self.get_values()
            if isinstance(other, (list, np.ndarray)):
                other = np.asarray(other)

            with np.errstate(all='ignore'):
                res = na_op(values, other)
            if is_scalar(res):
                raise TypeError('Could not compare %s type with Series' %
                                type(other))

            # always return a full value series here
            res = _values_from_object(res)

        res = pd.Series(res, index=self.index, name=self.name, dtype='bool')
        return res

    return wrapper


def _bool_method_SERIES(op, name, str_rep):
    """
    Wrapper function for Series arithmetic operations, to avoid
    code duplication.
    """

    def na_op(x, y):
        try:
            result = op(x, y)
        except TypeError:
            if isinstance(y, list):
                y = lib.list_to_object_array(y)

            if isinstance(y, (np.ndarray, ABCSeries)):
                if (is_bool_dtype(x.dtype) and is_bool_dtype(y.dtype)):
                    result = op(x, y)  # when would this be hit?
                else:
                    x = _ensure_object(x)
                    y = _ensure_object(y)
                    result = lib.vec_binop(x, y, op)
            else:
                try:

                    # let null fall thru
                    if not isnull(y):
                        y = bool(y)
                    result = lib.scalar_binop(x, y, op)
                except:
                    raise TypeError("cannot compare a dtyped [{0}] array with "
                                    "a scalar of type [{1}]".format(
                                        x.dtype, type(y).__name__))

        return result

    def wrapper(self, other):
        is_self_int_dtype = is_integer_dtype(self.dtype)

        fill_int = lambda x: x.fillna(0)
        fill_bool = lambda x: x.fillna(False).astype(bool)

        self, other = _align_method_SERIES(self, other, align_asobject=True)

        if isinstance(other, ABCSeries):
            name = _maybe_match_name(self, other)
            is_other_int_dtype = is_integer_dtype(other.dtype)
            other = fill_int(other) if is_other_int_dtype else fill_bool(other)

            filler = (fill_int if is_self_int_dtype and is_other_int_dtype
                      else fill_bool)
            return filler(self._constructor(na_op(self.values, other.values),
                                            index=self.index, name=name))

        elif isinstance(other, pd.DataFrame):
            return NotImplemented

        else:
            # scalars, list, tuple, np.array
            filler = (fill_int if is_self_int_dtype and
                      is_integer_dtype(np.asarray(other)) else fill_bool)
            return filler(self._constructor(
                na_op(self.values, other),
                index=self.index)).__finalize__(self)

    return wrapper


_op_descriptions = {'add': {'op': '+',
                            'desc': 'Addition',
                            'reversed': False,
                            'reverse': 'radd'},
                    'sub': {'op': '-',
                            'desc': 'Subtraction',
                            'reversed': False,
                            'reverse': 'rsub'},
                    'mul': {'op': '*',
                            'desc': 'Multiplication',
                            'reversed': False,
                            'reverse': 'rmul'},
                    'mod': {'op': '%',
                            'desc': 'Modulo',
                            'reversed': False,
                            'reverse': 'rmod'},
                    'pow': {'op': '**',
                            'desc': 'Exponential power',
                            'reversed': False,
                            'reverse': 'rpow'},
                    'truediv': {'op': '/',
                                'desc': 'Floating division',
                                'reversed': False,
                                'reverse': 'rtruediv'},
                    'floordiv': {'op': '//',
                                 'desc': 'Integer division',
                                 'reversed': False,
                                 'reverse': 'rfloordiv'},
                    'divmod': {'op': 'divmod',
                               'desc': 'Integer division and modulo',
                               'reversed': False,
                               'reverse': None},

                    'eq': {'op': '==',
                                 'desc': 'Equal to',
                                 'reversed': False,
                                 'reverse': None},
                    'ne': {'op': '!=',
                                 'desc': 'Not equal to',
                                 'reversed': False,
                                 'reverse': None},
                    'lt': {'op': '<',
                                 'desc': 'Less than',
                                 'reversed': False,
                                 'reverse': None},
                    'le': {'op': '<=',
                                 'desc': 'Less than or equal to',
                                 'reversed': False,
                                 'reverse': None},
                    'gt': {'op': '>',
                                 'desc': 'Greater than',
                                 'reversed': False,
                                 'reverse': None},
                    'ge': {'op': '>=',
                                 'desc': 'Greater than or equal to',
                                 'reversed': False,
                                 'reverse': None}}

_op_names = list(_op_descriptions.keys())
for k in _op_names:
    reverse_op = _op_descriptions[k]['reverse']
    _op_descriptions[reverse_op] = _op_descriptions[k].copy()
    _op_descriptions[reverse_op]['reversed'] = True
    _op_descriptions[reverse_op]['reverse'] = k


_flex_doc_SERIES = """
%s of series and other, element-wise (binary operator `%s`).

Equivalent to ``%s``, but with support to substitute a fill_value for
missing data in one of the inputs.

Parameters
----------
other : Series or scalar value
fill_value : None or float value, default None (NaN)
    Fill missing (NaN) values with this value. If both Series are
    missing, the result will be missing
level : int or name
    Broadcast across a level, matching Index values on the
    passed MultiIndex level

Returns
-------
result : Series

See also
--------
Series.%s
"""


def _flex_method_SERIES(op, name, str_rep, default_axis=None, fill_zeros=None,
                        **eval_kwargs):
    op_name = name.replace('__', '')
    op_desc = _op_descriptions[op_name]
    if op_desc['reversed']:
        equiv = 'other ' + op_desc['op'] + ' series'
    else:
        equiv = 'series ' + op_desc['op'] + ' other'

    doc = _flex_doc_SERIES % (op_desc['desc'], op_name, equiv,
                              op_desc['reverse'])

    @Appender(doc)
    def flex_wrapper(self, other, level=None, fill_value=None, axis=0):
        # validate axis
        if axis is not None:
            self._get_axis_number(axis)
        if isinstance(other, ABCSeries):
            return self._binop(other, op, level=level, fill_value=fill_value)
        elif isinstance(other, (np.ndarray, list, tuple)):
            if len(other) != len(self):
                raise ValueError('Lengths must be equal')
            return self._binop(self._constructor(other, self.index), op,
                               level=level, fill_value=fill_value)
        else:
            if fill_value is not None:
                self = self.fillna(fill_value)

            return self._constructor(op(self, other),
                                     self.index).__finalize__(self)

    flex_wrapper.__name__ = name
    return flex_wrapper


series_flex_funcs = dict(flex_arith_method=_flex_method_SERIES,
                         flex_comp_method=_flex_method_SERIES)

series_special_funcs = dict(arith_method=_arith_method_SERIES,
                            comp_method=_comp_method_SERIES,
                            bool_method=_bool_method_SERIES,
                            have_divmod=True)

_arith_doc_FRAME = """
Binary operator %s with support to substitute a fill_value for missing data in
one of the inputs

Parameters
----------
other : Series, DataFrame, or constant
axis : {0, 1, 'index', 'columns'}
    For Series input, axis to match Series index on
fill_value : None or float value, default None
    Fill missing (NaN) values with this value. If both DataFrame locations are
    missing, the result will be missing
level : int or name
    Broadcast across a level, matching Index values on the
    passed MultiIndex level

Notes
-----
Mismatched indices will be unioned together

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

_flex_doc_FRAME = """
%s of dataframe and other, element-wise (binary operator `%s`).

Equivalent to ``%s``, but with support to substitute a fill_value for
missing data in one of the inputs.

Parameters
----------
other : Series, DataFrame, or constant
axis : {0, 1, 'index', 'columns'}
    For Series input, axis to match Series index on
fill_value : None or float value, default None
    Fill missing (NaN) values with this value. If both DataFrame
    locations are missing, the result will be missing
level : int or name
    Broadcast across a level, matching Index values on the
    passed MultiIndex level

Notes
-----
Mismatched indices will be unioned together

Returns
-------
result : DataFrame

See also
--------
DataFrame.%s
"""


def _align_method_FRAME(left, right, axis):
    """ convert rhs to meet lhs dims if input is list, tuple or np.ndarray """

    def to_series(right):
        msg = 'Unable to coerce to Series, length must be {0}: given {1}'
        if axis is not None and left._get_axis_name(axis) == 'index':
            if len(left.index) != len(right):
                raise ValueError(msg.format(len(left.index), len(right)))
            right = left._constructor_sliced(right, index=left.index)
        else:
            if len(left.columns) != len(right):
                raise ValueError(msg.format(len(left.columns), len(right)))
            right = left._constructor_sliced(right, index=left.columns)
        return right

    if isinstance(right, (list, tuple)):
        right = to_series(right)

    elif isinstance(right, np.ndarray) and right.ndim:  # skips np scalar

        if right.ndim == 1:
            right = to_series(right)

        elif right.ndim == 2:
            if left.shape != right.shape:
                msg = ("Unable to coerce to DataFrame, "
                       "shape must be {0}: given {1}")
                raise ValueError(msg.format(left.shape, right.shape))

            right = left._constructor(right, index=left.index,
                                      columns=left.columns)
        else:
            msg = 'Unable to coerce to Series/DataFrame, dim must be <= 2: {0}'
            raise ValueError(msg.format(right.shape, ))

    return right


def _arith_method_FRAME(op, name, str_rep=None, default_axis='columns',
                        fill_zeros=None, **eval_kwargs):
    def na_op(x, y):
        try:
            result = expressions.evaluate(op, str_rep, x, y,
                                          raise_on_error=True, **eval_kwargs)
        except TypeError:
            xrav = x.ravel()
            if isinstance(y, (np.ndarray, ABCSeries)):
                dtype = np.find_common_type([x.dtype, y.dtype], [])
                result = np.empty(x.size, dtype=dtype)
                yrav = y.ravel()
                mask = notnull(xrav) & notnull(yrav)
                xrav = xrav[mask]

                # we may need to manually
                # broadcast a 1 element array
                if yrav.shape != mask.shape:
                    yrav = np.empty(mask.shape, dtype=yrav.dtype)
                    yrav.fill(yrav.item())

                yrav = yrav[mask]
                if np.prod(xrav.shape) and np.prod(yrav.shape):
                    with np.errstate(all='ignore'):
                        result[mask] = op(xrav, yrav)
            elif hasattr(x, 'size'):
                result = np.empty(x.size, dtype=x.dtype)
                mask = notnull(xrav)
                xrav = xrav[mask]
                if np.prod(xrav.shape):
                    with np.errstate(all='ignore'):
                        result[mask] = op(xrav, y)
            else:
                raise TypeError("cannot perform operation {op} between "
                                "objects of type {x} and {y}".format(
                                    op=name, x=type(x), y=type(y)))

            result, changed = maybe_upcast_putmask(result, ~mask, np.nan)
            result = result.reshape(x.shape)

        result = missing.fill_zeros(result, x, y, name, fill_zeros)

        return result

    if name in _op_descriptions:
        op_name = name.replace('__', '')
        op_desc = _op_descriptions[op_name]
        if op_desc['reversed']:
            equiv = 'other ' + op_desc['op'] + ' dataframe'
        else:
            equiv = 'dataframe ' + op_desc['op'] + ' other'

        doc = _flex_doc_FRAME % (op_desc['desc'], op_name, equiv,
                                 op_desc['reverse'])
    else:
        doc = _arith_doc_FRAME % name

    @Appender(doc)
    def f(self, other, axis=default_axis, level=None, fill_value=None):

        other = _align_method_FRAME(self, other, axis)

        if isinstance(other, pd.DataFrame):  # Another DataFrame
            return self._combine_frame(other, na_op, fill_value, level)
        elif isinstance(other, ABCSeries):
            return self._combine_series(other, na_op, fill_value, axis, level)
        else:
            if fill_value is not None:
                self = self.fillna(fill_value)

            return self._combine_const(other, na_op)

    f.__name__ = name

    return f


# Masker unused for now
def _flex_comp_method_FRAME(op, name, str_rep=None, default_axis='columns',
                            masker=False):
    def na_op(x, y):
        try:
            result = op(x, y)
        except TypeError:
            xrav = x.ravel()
            result = np.empty(x.size, dtype=bool)
            if isinstance(y, (np.ndarray, ABCSeries)):
                yrav = y.ravel()
                mask = notnull(xrav) & notnull(yrav)
                result[mask] = op(np.array(list(xrav[mask])),
                                  np.array(list(yrav[mask])))
            else:
                mask = notnull(xrav)
                result[mask] = op(np.array(list(xrav[mask])), y)

            if op == operator.ne:  # pragma: no cover
                np.putmask(result, ~mask, True)
            else:
                np.putmask(result, ~mask, False)
            result = result.reshape(x.shape)

        return result

    @Appender('Wrapper for flexible comparison methods %s' % name)
    def f(self, other, axis=default_axis, level=None):

        other = _align_method_FRAME(self, other, axis)

        if isinstance(other, pd.DataFrame):  # Another DataFrame
            return self._flex_compare_frame(other, na_op, str_rep, level)

        elif isinstance(other, ABCSeries):
            return self._combine_series(other, na_op, None, axis, level)
        else:
            return self._combine_const(other, na_op)

    f.__name__ = name

    return f


def _comp_method_FRAME(func, name, str_rep, masker=False):
    @Appender('Wrapper for comparison method %s' % name)
    def f(self, other):
        if isinstance(other, pd.DataFrame):  # Another DataFrame
            return self._compare_frame(other, func, str_rep)
        elif isinstance(other, ABCSeries):
            return self._combine_series_infer(other, func)
        else:

            # straight boolean comparisions we want to allow all columns
            # (regardless of dtype to pass thru) See #4537 for discussion.
            res = self._combine_const(other, func, raise_on_error=False)
            return res.fillna(True).astype(bool)

    f.__name__ = name

    return f


frame_flex_funcs = dict(flex_arith_method=_arith_method_FRAME,
                        flex_comp_method=_flex_comp_method_FRAME)

frame_special_funcs = dict(arith_method=_arith_method_FRAME,
                           comp_method=_comp_method_FRAME,
                           bool_method=_arith_method_FRAME)


def _arith_method_PANEL(op, name, str_rep=None, fill_zeros=None,
                        default_axis=None, **eval_kwargs):
    # copied from Series na_op above, but without unnecessary branch for
    # non-scalar
    def na_op(x, y):
        try:
            result = expressions.evaluate(op, str_rep, x, y,
                                          raise_on_error=True, **eval_kwargs)
        except TypeError:

            # TODO: might need to find_common_type here?
            result = np.empty(len(x), dtype=x.dtype)
            mask = notnull(x)
            result[mask] = op(x[mask], y)
            result, changed = maybe_upcast_putmask(result, ~mask, np.nan)

        result = missing.fill_zeros(result, x, y, name, fill_zeros)
        return result

    # work only for scalars
    def f(self, other):
        if not is_scalar(other):
            raise ValueError('Simple arithmetic with %s can only be '
                             'done with scalar values' %
                             self._constructor.__name__)

        return self._combine(other, op)

    f.__name__ = name
    return f


def _comp_method_PANEL(op, name, str_rep=None, masker=False):
    def na_op(x, y):
        try:
            result = expressions.evaluate(op, str_rep, x, y,
                                          raise_on_error=True)
        except TypeError:
            xrav = x.ravel()
            result = np.empty(x.size, dtype=bool)
            if isinstance(y, np.ndarray):
                yrav = y.ravel()
                mask = notnull(xrav) & notnull(yrav)
                result[mask] = op(np.array(list(xrav[mask])),
                                  np.array(list(yrav[mask])))
            else:
                mask = notnull(xrav)
                result[mask] = op(np.array(list(xrav[mask])), y)

            if op == operator.ne:  # pragma: no cover
                np.putmask(result, ~mask, True)
            else:
                np.putmask(result, ~mask, False)
            result = result.reshape(x.shape)

        return result

    @Appender('Wrapper for comparison method %s' % name)
    def f(self, other, axis=None):
        # Validate the axis parameter
        if axis is not None:
            axis = self._get_axis_number(axis)

        if isinstance(other, self._constructor):
            return self._compare_constructor(other, na_op)
        elif isinstance(other, (self._constructor_sliced, pd.DataFrame,
                                ABCSeries)):
            raise Exception("input needs alignment for this object [%s]" %
                            self._constructor)
        else:
            return self._combine_const(other, na_op)

    f.__name__ = name

    return f


panel_special_funcs = dict(arith_method=_arith_method_PANEL,
                           comp_method=_comp_method_PANEL,
                           bool_method=_arith_method_PANEL)