BUG: df.rename raises when multiindex has duplicates

[GYHHAHA]

>>> idx = pd.Index([('A','cat'),('B','cat'),('B','cat')])
>>> df = pd.DataFrame(index=idx)
>>> df.rename(index={'A':'Apple'}, level=0)
ValueError: non-broadcastable output operand with shape (1,1) doesn't match the broadcast shape (1,2)

[rhshadrach]

Thanks for the report @GYHHAHA! Further investigations and PRs are certainly welcome.

[GYHHAHA]

Hi, where this function tested? I think the error is related to this, but can't find test case.

pandas/pandas/_libs/index.pyx

Line 572 in 54f23e8

def _extract_level_codes(self, object target):

And also confused about _codes_to_ints func, any explanation about this bit computation? Thanks.

pandas/pandas/core/indexes/multi.py

Line 107 in 1cc030e

codes <<= self.offsets

[mroeschke]

It appears this work on master now, I think. Could use a test

In [1]: >>> idx = pd.Index([('A','cat'),('B','cat'),('B','cat')])
   ...: >>> df = pd.DataFrame(index=idx)
   ...: >>> df.rename(index={'A':'Apple'}, level=0)
Out[1]:
Empty DataFrame
Columns: []
Index: [(Apple, cat), (B, cat), (B, cat)]

[andrewwoood]

I was looking to take this on as my first open source contribution, but when I ran the code locally I'm running into the error mentioned initially:

>>> idx = pd.Index([('A','cat'),('B','cat'),('B','cat')])
>>> df = pd.DataFrame(index=idx)
>>> df.rename(index={'A':'Apple'}, level=0)
ValueError: non-broadcastable output operand with shape (1,1) doesn't match the broadcast shape (1,2)

In the case that I'm doing this incorrectly, could I take this issue on? @mroeschke, when you mention that it needs tests, does that mean running it locally with different values and verifying correctness? Or are there unit tests that I need to run somewhere? I'm not so clear on the process, sorry for asking so many questions.

[mroeschke]

@andrewwoood I was running this on the master branch, and it appeared to have worked. This code snippet would need to be added as a unit test in the pandas testing suite.

[andrewwoood]

take