Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

API: Add equals method to NDFrames. #5283

merged 1 commit into from Jan 24, 2014
Changes from all commits
File filter...
Filter file types
Jump to…
Jump to file or symbol
Failed to load files and symbols.


Just for now

@@ -215,6 +215,14 @@ These operations produce a pandas object the same type as the left-hand-side inp
that if of dtype ``bool``. These ``boolean`` objects can be used in indexing operations,
see :ref:`here<indexing.boolean>`

As of v0.13.1, Series, DataFrames and Panels have an equals method to compare if

This comment has been minimized.

Copy link

jreback Jan 24, 2014


I merged this thanks! maybe as a small followup....can you explain in the docs why one would need to do this, maybe a small example is in order?

two such objects are equal.

.. ipython:: python
.. _basics.reductions:

Boolean Reductions
@@ -66,6 +66,7 @@ API Changes
timedeltas (:issue:`5458`,:issue:`5689`)
- Add ``-NaN`` and ``-nan`` to the default set of NA values
(:issue:`5952`). See :ref:`NA Values <io.na_values>`.
- ``NDFrame`` now has an ``equals`` method. (:issue:`5283`)

Experimental Features
@@ -42,6 +42,21 @@ API changes

- Add ``-NaN`` and ``-nan`` to the default set of NA values (:issue:`5952`).
See :ref:`NA Values <io.na_values>`.
- Added the ``NDFrame.equals()`` method to compare if two NDFrames are
equal have equal axes, dtypes, and values. Added the
``array_equivalent`` function to compare if two ndarrays are
equal. NaNs in identical locations are treated as
equal. (:issue:`5283`)

.. ipython:: python

df = DataFrame({'col':['foo', 0, np.nan]}).sort()
df2 = DataFrame({'col':[np.nan, 0, 'foo']}, index=[2,1,0])

import pandas.core.common as com
com.array_equivalent(np.array([0, np.nan]), np.array([0, np.nan]))
np.array_equal(np.array([0, np.nan]), np.array([0, np.nan]))

Prior Version Deprecations/Changes
@@ -277,6 +277,42 @@ def notnull(obj):
return -res

def array_equivalent(left, right):

This comment has been minimized.

Copy link

unutbu Jan 21, 2014

Author Contributor

NDFrame.equals is now implemented (in without using array_equivalent. Should array_equivalent be moved to pandas/util/

This comment has been minimized.

Copy link

jreback Jan 21, 2014


I think the goal was to use this (eventually?) in Index comparisons as well IIRC. (I believe their were some perf issues in that regards to not doing that now?) In any event let's defer using it in index for now. @jtratner ?

array_equivalent is defacto np.array_equal but handles nan comparisons, yes?

True if two arrays, left and right, have equal non-NaN elements, and NaNs in
corresponding locations. False otherwise. It is assumed that left and right
are NumPy arrays of the same dtype. The behavior of this function
(particularly with respect to NaNs) is not defined if the dtypes are
left, right : array_like
Input arrays.
b : bool
Returns True if the arrays are equivalent.
>>> array_equivalent([1, 2, nan], np.array([1, 2, nan]))
>>> array_equivalent([1, nan, 2], [1, 2, nan])
if left.shape != right.shape: return False
# NaNs occur only in object arrays, float or complex arrays.
if left.dtype == np.object_:
# If object array, we need to use pd.isnull
return ((left == right) | pd.isnull(left) & pd.isnull(right)).all()
elif not issubclass(left.dtype.type, (np.floating, np.complexfloating)):
# if not a float or complex array, then there are no NaNs
return np.array_equal(left, right)
# For float or complex arrays, using np.isnan is faster than pd.isnull
return ((left == right) | (np.isnan(left) & np.isnan(right))).all()

def _iterable_not_string(x):
return (isinstance(x, collections.Iterable) and
not isinstance(x, compat.string_types))
@@ -608,6 +608,15 @@ def __invert__(self):
arr = operator.inv(_values_from_object(self))
return self._wrap_array(arr, self.axes, copy=False)

def equals(self, other):
Determines if two NDFrame objects contain the same elements. NaNs in the
same location are considered equal.
if not isinstance(other, self._constructor):
return False
return self._data.equals(other._data)

# Iteration

@@ -1121,13 +1121,24 @@ def func(c, v, o):

return result_blocks

def equals(self, other):
if self.dtype != other.dtype or self.shape != other.shape: return False
return np.array_equal(self.values, other.values)

This comment has been minimized.

Copy link

unutbu Jan 21, 2014

Author Contributor

I just want to record this here, since my thoughts on this matter have been so muddled:

np.array_equal treats NaNs in object arrays as equal:

>>> np.array_equal(np.array([np.nan], dtype='O'), np.array([np.nan], dtype='O'))

so object arrays do not need to be special-cased, ObjectBlock.equals does not need to be defined, and ObjectBlock.equals can defer to Block.equals. Only FloatBlocks and ComplexBlocks need to define their own equals method, since, for example,

>>> np.array_equal(np.array([np.nan], dtype='<f4'), np.array([np.nan], dtype='<f4'))
>>> np.array_equal(np.array([np.nan], dtype='complex128'), np.array([np.nan], dtype='complex128'))

Thankfully, np.array_equal also treats NaTs as equal:

>>> np.array_equal(np.array([np.datetime64('nat')]), np.array([np.datetime64('nat')]))

NaNs can not exist in bool arrays:

>>> np.array([np.nan], dtype='bool')
array([ True], dtype=bool)

or can NaNs exist in timedeltas:

>>> np.timedelta64(np.nan,'D')
ValueError: Could not convert object to NumPy timedelta

so np.array_equal should work as usual for these kinds of values.

class NumericBlock(Block):
is_numeric = True
_can_hold_na = True

class FloatBlock(NumericBlock):
class FloatOrComplexBlock(NumericBlock):
def equals(self, other):
if self.dtype != other.dtype or self.shape != other.shape: return False
left, right = self.values, other.values
return ((left == right) | (np.isnan(left) & np.isnan(right))).all()

class FloatBlock(FloatOrComplexBlock):
is_float = True
_downcast_dtype = 'int64'

@@ -1166,8 +1177,7 @@ def should_store(self, value):
return (issubclass(value.dtype.type, np.floating) and
value.dtype == self.dtype)

class ComplexBlock(NumericBlock):
class ComplexBlock(FloatOrComplexBlock):
is_complex = True

def _can_hold_element(self, element):
@@ -2563,7 +2573,7 @@ def get_data(self, copy=False, columns=None, **kwargs):
return self.combine(blocks)

def combine(self, blocks):
""" reutrn a new manager with the blocks """
""" return a new manager with the blocks """
indexer = np.sort(np.concatenate([b.ref_locs for b in blocks]))
new_items = self.items.take(indexer)

@@ -3491,6 +3501,16 @@ def item_dtypes(self):
raise AssertionError('Some items were not in any block')
return result

def equals(self, other):
self_axes, other_axes = self.axes, other.axes
if len(self_axes) != len(other_axes):
return False
if not all (ax1.equals(ax2) for ax1, ax2 in zip(self_axes, other_axes)):
return False
return all(block.equals(oblock) for block, oblock in
zip(self.blocks, other.blocks))

class SingleBlockManager(BlockManager):

@@ -4004,6 +4024,9 @@ def _merge_blocks(blocks, items, dtype=None, _can_consolidate=True):
raise AssertionError("_merge_blocks are invalid!")
dtype = blocks[0].dtype

if not items.is_unique:

This comment has been minimized.

Copy link

unutbu Jan 23, 2014

Author Contributor

The example you gave did indeed break the code. I've added your example to and am handling this case by sorting the blocks according to their ref_locs.

blocks = sorted(blocks, key=lambda b: b.ref_locs.tolist())

new_values = _vstack([b.values for b in blocks], dtype)
new_items = blocks[0].items.append([b.items for b in blocks[1:]])
new_block = make_block(new_values, new_items, items)
@@ -5,11 +5,10 @@
from import assert_equal
import numpy as np
from pandas.tslib import iNaT, NaT

from pandas import Series, DataFrame, date_range, DatetimeIndex, Timestamp
from pandas import compat
from pandas.compat import range, long, lrange, lmap, u
from pandas.core.common import notnull, isnull
from pandas.core.common import notnull, isnull, array_equivalent
import pandas.core.common as com
import pandas.util.testing as tm
import pandas.core.config as cf
@@ -167,6 +166,19 @@ def test_downcast_conv():
result = com._possibly_downcast_to_dtype(arr,'infer')
tm.assert_almost_equal(result, expected)

def test_array_equivalent():

This comment has been minimized.

Copy link

unutbu Jan 21, 2014

Author Contributor

If array_equivalent is moved to pandas/util/ then I'll move this to pandas/tests/

assert array_equivalent(np.array([np.nan, np.nan]), np.array([np.nan, np.nan]))
assert array_equivalent(np.array([np.nan, 1, np.nan]), np.array([np.nan, 1, np.nan]))
assert array_equivalent(np.array([np.nan, None], dtype='object'),
np.array([np.nan, None], dtype='object'))
assert array_equivalent(np.array([np.nan, 1+1j], dtype='complex'),
np.array([np.nan, 1+1j], dtype='complex'))
assert not array_equivalent(np.array([np.nan, 1+1j], dtype='complex'),
np.array([np.nan, 1+2j], dtype='complex'))
assert not array_equivalent(np.array([np.nan, 1, np.nan]), np.array([np.nan, 2, np.nan]))
assert not array_equivalent(np.array(['a', 'b', 'c', 'd']), np.array(['e', 'e']))

def test_datetimeindex_from_empty_datetime64_array():
for unit in [ 'ms', 'us', 'ns' ]:
idx = DatetimeIndex(np.array([], dtype='datetime64[%s]' % unit))
@@ -812,7 +812,8 @@ def test_metadata_propagation_indiv(self):

# resample
df = DataFrame(np.random.randn(1000,2), index=date_range('20130101',periods=1000,freq='s'))
df = DataFrame(np.random.randn(1000,2),
result = df.resample('1T')

@@ -851,6 +852,80 @@ def test_squeeze(self):
p4d = tm.makePanel4D().reindex(labels=['label1'],items=['ItemA'])

def test_equals(self):
s1 = pd.Series([1, 2, 3], index=[0, 2, 1])
s2 = s1.copy()

s1[1] = 99
self.assert_(not s1.equals(s2))

# NaNs compare as equal
s1 = pd.Series([1, np.nan, 3, np.nan], index=[0, 2, 1, 3])
s2 = s1.copy()

s2[0] = 9.9
self.assert_(not s1.equals(s2))

idx = MultiIndex.from_tuples([(0, 'a'), (1, 'b'), (2, 'c')])
s1 = Series([1, 2, np.nan], index=idx)
s2 = s1.copy()

# Add object dtype column with nans
index = np.random.random(10)
df1 = DataFrame(np.random.random(10,), index=index, columns=['floats'])
df1['text'] = 'the sky is so blue. we could use more chocolate.'.split()
df1['start'] = date_range('2000-1-1', periods=10, freq='T')
df1['end'] = date_range('2000-1-1', periods=10, freq='D')
df1['diff'] = df1['end'] - df1['start']
df1['bool'] = (np.arange(10) % 3 == 0)
df1.ix[::2] = nan
df2 = df1.copy()
self.assert_(not df1.equals(object))

# different dtype
different = df1.copy()
different['floats'] = different['floats'].astype('float32')
self.assert_(not df1.equals(different))

# different index
different_index = -index
different = df2.set_index(different_index)
self.assert_(not df1.equals(different))

# different columns
different = df2.copy()
different.columns = df2.columns[::-1]
self.assert_(not df1.equals(different))

# DatetimeIndex
index = pd.date_range('2000-1-1', periods=10, freq='T')
df1 = df1.set_index(index)
df2 = df1.copy()

# MultiIndex
df3 = df1.set_index(['text'], append=True)
df2 = df1.set_index(['text'], append=True)

df2 = df1.set_index(['floats'], append=True)
self.assert_(not df3.equals(df2))

# NaN in index
df3 = df1.set_index(['floats'], append=True)
df2 = df1.set_index(['floats'], append=True)

if __name__ == '__main__':
nose.runmodule(argv=[__file__, '-vvs', '-x', '--pdb', '--pdb-failure'],
@@ -584,6 +584,27 @@ def test_missing_unicode_key(self):
except KeyError:
pass # this is the expected exception

def test_equals(self):
# unique items
index = Index(list('abcdef'))
block1 = make_block(np.arange(12).reshape(3,4), list('abc'), index)
block2 = make_block(np.arange(12).reshape(3,4)*10, list('def'), index)
block1.ref_items = block2.ref_items = index
bm1 = BlockManager([block1, block2], [index, np.arange(block1.shape[1])])
bm2 = BlockManager([block2, block1], [index, np.arange(block1.shape[1])])

# non-unique items
index = Index(list('aaabbb'))
block1 = make_block(np.arange(12).reshape(3,4), list('aaa'), index,
block2 = make_block(np.arange(12).reshape(3,4)*10, list('bbb'), index,
block1.ref_items = block2.ref_items = index
bm1 = BlockManager([block1, block2], [index, np.arange(block1.shape[1])])
bm2 = BlockManager([block2, block1], [index, np.arange(block1.shape[1])])

if __name__ == '__main__':
import nose
nose.runmodule(argv=[__file__, '-vvs', '-x', '--pdb', '--pdb-failure'],
@@ -436,6 +436,7 @@ def isiterable(obj):
def is_sorted(seq):
return assert_almost_equal(seq, np.sort(np.array(seq)))

# This could be refactored to use the NDFrame.equals method
def assert_series_equal(left, right, check_dtype=True,
@@ -455,7 +456,7 @@ def assert_series_equal(left, right, check_dtype=True,
assert_attr_equal('dtype', left.index, right.index)
assert_attr_equal('inferred_type', left.index, right.index)

# This could be refactored to use the NDFrame.equals method
def assert_frame_equal(left, right, check_dtype=True,
@@ -370,3 +370,36 @@ def f(K=100):
frame_dtypes = Benchmark('df.dtypes', setup,

# equals
setup = common_setup + """
def make_pair(name):
df = globals()[name]
df2 = df.copy()
df2.ix[-1,-1] = np.nan
return df, df2
def test_equal(name):
df, df2 = pairs[name]
return df.equals(df)
def test_unequal(name):
df, df2 = pairs[name]
return df.equals(df2)
float_df = DataFrame(np.random.randn(1000, 1000))
object_df = DataFrame([['foo']*1000]*1000)
nonunique_cols = object_df.copy()
nonunique_cols.columns = ['A']*len(nonunique_cols.columns)
pairs = dict([(name,make_pair(name))
for name in ('float_df', 'object_df', 'nonunique_cols')])
frame_float_equal = Benchmark('test_equal("float_df")', setup)
frame_object_equal = Benchmark('test_equal("object_df")', setup)
frame_nonunique_equal = Benchmark('test_equal("nonunique_cols")', setup)

frame_float_unequal = Benchmark('test_unequal("float_df")', setup)
frame_object_unequal = Benchmark('test_unequal("object_df")', setup)
frame_nonunique_unequal = Benchmark('test_unequal("nonunique_cols")', setup)

ProTip! Use n and p to navigate between commits in a pull request.
You can’t perform that action at this time.