Relax pandas<3 pin#4691
Conversation
h-mayorquin
left a comment
h-mayorquin
left a comment
There was a problem hiding this comment.
LGTM
It is is a breaking change or users who for some reason can't use the older pandas but I guess they have older versions of SpikeInterface for that,
|
I'm not sure that dropping 3.10 support is worth it to support pandas 3 - I think we should at least discuss it. This would leave us only supporting three versions of Python. (I'm ok with the PR from a technical standpoint though!) |
|
I agree with Chris. 3.10 isn't EOL until October so removing support before then at least seems premature to me. |
Both pandas 3 and zarr 3 dropped support for 3.10. I don't think this will be an issue. Hopefully zarr v3 will be in main by then too |
And I agree with @h-mayorquin here too. If someone wants to use 3.10, we have plenty of older versions (including the current one) with support :) |
|
Plus, we will support 3.14 by next release (#4173 is in milestone 0.105.0) |
|
Tested locally, and none of the pandas 3 changes we used are breaking. So we don't need to pin pandas, and hence don't need to remove 3.10 support. Note: The 3.10 tests are using pandas<3 and the 3.13 tests are using pandas>3, as expected |
| if upset_data.index.nlevels == 1: | ||
| warnings.warn( | ||
| f"Skipping UpSet plot: only one metric, {memberships[0][0]}, was responsible for all failures." | ||
| ) | ||
| continue | ||
|
|
There was a problem hiding this comment.
UpSet plot requires there to be at least two metrics which cause "failures" (units != sua). Sometimes, only amplitude_cutoff was causing all failures, meaning it couldn't plot and would error.
chrishalcrow
changed the title
4 participants
Fixes #4687