Partial string matching for timestamps with multiindex

[thrasibule]

I'm trying to get a slice from a multiindex. I find the behavior pretty inconsistent. Here is a simple dataframe:

test = pd.DataFrame({'l': ['A', 'B', 'C', 'C', 'B', 'A'], 'm': [pd.Timestamp('2014-06-11 14:26:27'), pd.Timestamp('2014-06-11 15:26:27'),  pd.Timestamp('2014-06-11 16:26:27'), pd.Timestamp('2014-06-12 14:26:27'), pd.Timestamp('2014-06-12 15:26:27'),  pd.Timestamp('2014-06-12 16:26:27')]})

With a single index, I can select all the data for a given day as follows:

test_single_index = test.set_index(['m'])
#partial string indexing works
test_single_index.loc['2014-06-11']

But it doesn't work for a multiindex:

test_multi_index = test.set_index(['m', 'l'])
#would expect this to work
test_multi_index.loc['2014-06-11']
#or this
test_multi_index.loc[('2014-06-11', 'A'),:]

#what works
idx = pd.IndexSlice
test_multi_index.loc[idx['2014-06-11':'2014-06-11',:],:]
test_multi_index.loc[idx['2014-06-11','A'],:]

So I can make it work if I specify the slice explicitely, but it would be nice if the behavior for the 1D index carried over to Multiindices.

[jreback]

this is almost a dupe of #9732
This is simply not implemented (though not that difficult). See that issue for an easy work-around.

[markroth8]

Added some unit tests that should pass once this feature works. @jreback did a quick review pass on the tests.

@jreback, you said you'd have a look at the code path and give some recommendations on how to fix this issue.

[markroth8]

@jreback, per our discussion, I have an implementation that seems to fix this issue (in the private branch above), but it does not fail the test for df.loc[('2016-01-01', 'a'), :] which you thought it should fail. Before I make a formal pull request, can you have a look at the branch?

These tests all pass:
$ nosetests -A "not slow" tests/frame tests/series tests/indexes tests/test_indexing.py

[markroth8]

Additionally, df_swap.loc[idx[:, '2016-01-02':], :] fails with an error, but I believe this can be considered a separate bug, as it did so before my proposed fix.

The cause is that the code in multi.py is trying to call .stop on an int instead of first checking to see if stop is a slice:

if isinstance(start, slice) or isinstance(stop, slice):
                # we have a slice for start and/or stop
                # a partial date slicer on a DatetimeIndex generates a slice
                # note that the stop ALREADY includes the stopped point (if
                # it was a string sliced)
                return convert_indexer(start.start, stop.stop, step)

Please let me know how you'd like me to proceed with the pull request.