Keep name after DataFrame drop (issue 2939)

[hayd]

This should fix #2939.

[hayd]

Ag, although this is working locally, there is a typo in my test case... and yet it still passes.

Presumably this is because assert_frame_equal doesn't check the names (?), I'll update with an additional test for the names.

[hayd]

I removed the assert_frame_equal and only test the (index and column) names i.e. just what's been fixed.

(Aside: should assert_frame_equal test column/index names?)

[hayd]

Sorry, this last commit is a bad one, I shouldn't have done it. We shouldn't be checking names (only) on the check_type flag, as such this is only passing the tests by fluke. If I check all assert_frame_equalss with a check_names flag (i.e. always check names) I get a ton of errors. If I default it to False it would probably ok (but then it's not checking anything as nowhether uses check_name :s).

I should have added this to assert_frame_equal in a separate pull request, or at least waiting til your answer.

Is there a way to drop this last commit? (Should I just add another reverting it?)

[jreback]

http://stackoverflow.com/questions/927358/how-to-undo-last-git-commit
basically just

git reset HEAD-1

then u can recommit

rebase

then force push to update the remote

[jreback]

just add as an option to asset_frame_equal
set to False by default

[jreback]

or try getattr(left,'names',None) == getattr(right,'names',None)

so will check on if one side is bad

neither or both is ok though (as names prob
not propagated correctly in most situations)

or maybe it's just a few - not sure

[hayd]

Thanks @jreback, I will close this pull-request and add as separate pull.

[jreback]

ok

u can do it with this one ..

make a new commit that fixes what u want
git rebase -i commit_sha_that_starts_this_pr
git push myfork thisbranch -f

will replace the commits here with the revised ones

[hayd]

@jreback I'm not sure what you mean about getattr...

I squashed my commits to one, and added them, will make the pull-request now (I default check_names as False).

I'll have a play in a separate branch for assert_frame_equal to see where names isn't propogated, but I think it's a fair few places (whether it should be I'm not sure)...

I'd prefer to be able to set default check_names to True (but might be too much work to make that work), but doing so could be this throws up some current bugs similar to this one.

[hayd]

@jreback I'll give that a try for good measure :)

[jreback]

getattr will allow u to check names
but If the attribute does exist it will be none
so will compare as try on both sides (and test won't fail)

not sure how much these names are correctly propagating (or not)
so need to figure out what's appropriate

[hayd]

@jreback But it should always have the attribute names? I thought the reason where it's failing is because it isn't propogating. ?

I will investigate :)

[jreback]

I don't know - not sure when they r assigned or if a default is assigned
so maybe the attribute doesn't exist unless a name is created - worth having a look into the general case

[hayd]

I think it's always there:

In [1]: df = pd.DataFrame([])

In [2]: df.index.names
Out[2]: [None]

In [3]: df.columns.names
Out[3]: [None]

[jreback]

so if you set check_names to True
look and see what is failing and why - prob similar to the bug u fixed
something is not propagating

[hayd]

Precisely, so far it seems to be mostly expected results not including the name field. (I'm working through it.)

[hayd]

@jreback FYI I added the check_names=True as a new pull-request #2964.

I itemized the possible bugs/questionable behaviour there to do with index/column names, there's quite a couple I wasn't sure about (before I go "fixing" things).