Fix for issue #734

[kevin-duclos]

Change traj.set_index call to avoid Pandas error: 'ValueError: cannot reindex on an axis with duplicate labels'.

[caspervdw]

Thanks for your PR @kevin-duclos. I haven't worked with pandas for a while so I am not sure why this error is happening. But looking at the code, I think it might be a fix for your specific issue, but in general, the code should work for inputs without the "particle" column too. This is not reflected by the unittests however (they all use inputs with the "particle" column).

Maybe @nkeim can comment on the specific intended usage of this function?

[nkeim]

Thanks @kevin-duclos . The error in #734 makes sense, except it should occur any time that there are many particles, including in the trackpy tests. So I'm confused.

I made changes to subtract_drift() and its docstring that I think address @caspervdw 's concern. However, at least temporarily, I added some lines to the tests that are supposed to trigger this error. We'll see if any of the CI environments catch it.

[nkeim]

OK, I think this is ready to merge. In addition to the fix by @kevin-duclos , we now include two tests with no particle column:

• DataFrame with only one particle. This is unlikely to cause any reindexing problem.
• DataFrame with many (unlabeled) particles. This would be the case if one tried to use subtract_drift() with feature data prior to tracking. Note that compute_drift() requires a particle column, so the user would need to compute the drift some other way. I still don't know why this doesn't cause an error, but since it's a reasonable use case, we might as well test it.

I think this is ready to merge. @caspervdw does this address your question? @kevin-duclos do you think this still fixes the bug?

[nkeim]

Just bumping this issue, especially for @kevin-duclos : Is this OK to merge?

[kevin-duclos]

All good with me, @nkeim. Thanks!

[xyc2718]

After the 0.6.2 update, the dataframe processed by subtract_drift() poses an issue for imsd() due to the presence of "particle" in both row and column indices. To address this, it needs to use dataframe.reset_index(drop=True) before running the function to ensure proper execution.