ENH: NDFrame.equals

[jreback]

seems we need a public version of tm.assert_frame_equal

maybe equals on generic NDFrame objects

http://stackoverflow.com/questions/19322506/pandas-nan-comparison

seems good on the values comparison (should compare axes/types as well of course)

((df1 == df2) | ((df != df) & (df2 != df2))).values.all()

[jtratner]

Well, there's eq, lt, etc now on NDFrame, why not allow argument to eq
called something like fillna=True or comparena=True. so it'd just be
NDFrame.eq(fillna=True).all()

This would also be a good opportunity for a recursive=True kwarg to all
that lets you always get to a bool regardless of the NDFrame type you have.
(so all() for Series, all().all() for DataFrame and I think
all().all().all() for Panel)

seems we need a public version of tm.assert_frame_equal

maybe equals on generic NDFrame objects

http://stackoverflow.com/questions/19322506/pandas-nan-comparison

—
Reply to this email directly or view it on
GitHubhttps://github.com//issues/5183
.

[jreback]

eq and such are element wise operators

equals would signify an object comparator

[jtratner]

okay, that's reasonable, so it's just what you have + df1.index.equals(df2.index)? I still hanker to add a recursive all :P (plus you then would not need to copy to values).

Also, would it make more sense to do:

df1 = df1.values
df2 = df2.values
((df1 == df2) | ((df1 != df1) & (df2 != df2))).all()

instead? seems like that could be simpler. Also, I'm assuming you'd want to check if their shapes are compatible or even just the same (so that you can always safely use equals), so maybe: check shape (len(columns), len(index)), then check dtypes, etc. And then would you want to have a check_metadata argument too? (i.e., column names, index names, etc)

[jreback]

yep in fact these might be a nice replacement for assert_frame_equal (and it's friends)
and should be much faster too

[jtratner]

what about non-comparable things?

[jreback]

I think they should return False
eg series.equals(frame)

[jtratner]

and gotta check for matching shape first b/c of broadcasting

I'm a little skeptical on the speed benefit for tests, when I profiled things 6 months ago
I came up with <10% improvement potential. Small potatoes compared
to all the network tests that have found their way into the "fast" test suite.

I may have been wrong though.

[jreback]

@y-p oh...that's not really the reason for this (that's a side benefit)...

its the issue of comparing two objects that have nan's in them where if the nan locations match then it should be True (and not False)....so typical x == y fail on this

Fair enough, just apropos #3150.