Rows duplication with join_overlaps

[Picani]

Hi

First, thanks a lot for that incredible lib!

I encounter a bug while using join_overlaps in left-join fashion. When using an index in the left pandas DataFrame, if the index doesn't start at 0, one row is duplicated.

Examples

Without index

left = pd.DataFrame([
    {"Chromosome": "1", "Start": 100, "End": 200},
    {"Chromosome": "2", "Start": 300, "End": 400},
    
])

right = pd.DataFrame([
    {"Chromosome": "1", "Start": 150, "End": 160},
])

print(pr.PyRanges(left).join_overlaps(pr.PyRanges(right), join_type="left"))

Which prints as expected

index  |      Chromosome    Start      End    Start_b      End_b
int64  |             str    int64    int64    float64    float64
-------  ---  ------------  -------  -------  ---------  ---------
    0  |               1      100      200        150        160
    1  |               2      300      400        nan        nan
PyRanges with 2 rows, 5 columns, and 1 index columns.
Contains 2 chromosomes.

With an index starting at zero

left = pd.DataFrame([
    {"Chromosome": "1", "Start": 100, "End": 200},
    {"Chromosome": "2", "Start": 300, "End": 400},
], index=[0, 1])

right = pd.DataFrame([
    {"Chromosome": "1", "Start": 150, "End": 160},
])

print(pr.PyRanges(left).join_overlaps(pr.PyRanges(right), join_type="left"))

Which prints as expected

index  |      Chromosome    Start      End    Start_b      End_b
int64  |             str    int64    int64    float64    float64
-------  ---  ------------  -------  -------  ---------  ---------
    0  |               1      100      200        150        160
    1  |               2      300      400        nan        nan
PyRanges with 2 rows, 5 columns, and 1 index columns.
Contains 2 chromosomes.

Note that with [1, 0] as index, it works well too.

With an index starting at 1

left = pd.DataFrame([
    {"Chromosome": "1", "Start": 100, "End": 200},
    {"Chromosome": "2", "Start": 300, "End": 400},
], index=[1, 2])

right = pd.DataFrame([
    {"Chromosome": "1", "Start": 150, "End": 160},
])

print(pr.PyRanges(left).join_overlaps(pr.PyRanges(right), join_type="left"))

Which prints

index  |      Chromosome    Start      End    Start_b      End_b
int64  |             str    int64    int64    float64    float64
-------  ---  ------------  -------  -------  ---------  ---------
    0  |               1      100      200        150        160
    1  |               1      100      200        nan        nan
    2  |               2      300      400        nan        nan
PyRanges with 3 rows, 5 columns, and 1 index columns.
Contains 2 chromosomes.

Note that with [2, 1] as index, it gives the same wrong result.

Environment

• Python v3.12.10
• pandas v3.0.3
• pyranges1 v1.3.8

[endrebak]

Thanks for reporting. I wrote the function ages ago, never used it and I might have written it too hastily. Would be fun to see whether an LLM will be able to find the bug purely based on your report.

It seems like even if a left row contains an overlap it is also additionally reported as if it does not.

[endrebak]

This was due to me forgetting to update these functions after I changed all pyranges functions to return the original indices.

[Picani]

Thanks a lot for the quick fix!

[endrebak]

Thank you for reporting!