merge_asof(left_index=True, right_on=...) overwrites left index with right index

[gtg944q]

Building from an example in the docs...

left = pd.DataFrame({'left_val': ['a', 'b', 'c'], 'left_time': [1, 5, 10]}, index=[1, 5, 10])
right = pd.DataFrame({'right_val': [1, 2, 3, 6, 7]}, index=[1, 2, 3, 6, 7])

>>> pd.merge_asof(left, right, left_index=True, right_index=True)
   left_val  left_time  right_val
1         a          1          1
5         b          5          3
10        c         10          7

Returned index is the index of left.

>>> pd.merge_asof(left, right, left_on='left_time', right_index=True)
   left_val  left_time  right_val
1         a          1          1
5         b          5          3
10        c         10          7

Returned index is the index of left.

>>> pd.merge_asof(left, right, left_on='left_time', right_on='right_val')
  left_val  left_time  right_val
0        a          1          1
1        b          5          3
2        c         10          7

Return value has index reset (I assume expected?)

>>> pd.merge_asof(left, right, left_index=True, right_on='right_val')
  left_val  left_time  right_val
1        a          1          1
3        b          5          3
7        c         10          7

Returned index is the index of right.

Perhaps I'm misunderstanding something about how this is supposed to work, but it seems to me that the behavior in the last example is incorrect.

[gtg944q]

NB: proving to myself that the indexes returned in examples 2 and 4 are indeed the indexes, not the values:

>>> left = pd.DataFrame({'left_val': ['a', 'b', 'c'], 'left_time': [1, 4, 10]}, index=[1, 5, 10])
>>> pd.merge_asof(left, right, left_on='left_time', right_index=True)
   left_val  left_time  right_val
1         a          1          1
5         b          4          3
10        c         10          7
>>> right = pd.DataFrame({'right_val': [1, 2, 3, 6, 8]}, index=[1, 2, 3, 6, 7])
>>> pd.merge_asof(left, right, left_index=True, right_on='right_val')
  left_val  left_time  right_val
1        a          1          1
3        b          4          3
7        c         10          8

[ShaharNaveh]

@gtg944q Thank you for the report!

---

Do you mind paste the URL of the examples you found in the docs? we fixed a lot of documentation related problems recently, maybe this is already solved.

[gtg944q]

It's the second block of examples at https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.merge_asof.html, but I'm really just building on an example in there, so it's not a documentation problem.

[mroeschke]

Since this is advertised as "left-join" like, the last 2 examples you provide are surprising, and I'd expect to propagate the values in the left index (and not reset the index or use the right index). I'll mark as a bug, but this may break some existing tests.