Expose save preprocessing in ks4

[JoeZiminski]

This PR edits kilosort4.py to expose the save_preprocessed_copy parameter. Now, temp_wh.dat is saveed to the sorting output and params.py is edited to point phy to the temp_wh.dat. @carlacodes if you have some time maybe you could check off this fork/branch this is behaving as expected? I tested with the below code:

Test code

import numpy as np
import spikeinterface.full as si

recording, _, sorting = si.generate_drifting_recording(
    duration=10.0,
    generate_displacement_vector_kwargs=dict(
        displacement_sampling_frequency=5.0,
        drift_start_um=[0, 20],
        drift_stop_um=[0, -20],
        drift_step_um=1,
        motion_list=[
            dict(
                drift_mode="zigzag",
                non_rigid_gradient=None,
                t_start_drift=None,
                t_end_drift=None,
                period_s=200,
            ),
        ],
    ),
)

si.run_sorter("kilosort4", recording, save_preprocessed_copy=True)

At present this is not tested in #3085 but that can be extended after this PR is merged.

More generally @alejoe91, at present kilosort4.py replicates the body of run_sorter.py but as versions progress, this might prove difficult to maintain and requires a lot of testing e.g. #3085 to check nothing was missed. I wonder if long-term (assuming I am correct in thinking this is done to skip: preprocessing and drift correction), we can ask / work with KS4 team to expose options to skip preprocessing if they are interested. Then, we could just wrap the run_kilosort function directly and testing would just be checking the kwargs are passed properly (rather than having to check the outputs). Maybe there are other considerations I have overlooked though, WDYT?

[alejoe91]

@JoeZiminski this LGTM. @carlacodes were you able to test this?

[carlacodes]

LGTM; tested on my own data with ks 4.0.12

[zm711]

No way..... I have a feeling is is going to be fixed and then we will need another version just to account for this....

[alejoe91]

@zm711 you need to reviow KS PRs ahahahah

[dhmjhu]

It's fixed as of 4.0.13 (MouseLand/Kilosort@633ee92).

[alejoe91]

@JoeZiminski maybe we have to go back to positional arguments here? :)

[JoeZiminski]

Yes I think so @alejoe91 or the number of conditionals will explode, shall we merge this and then continue the discussion on how best to proceed here?

[alejoe91]

Let's do that!

[dhmjhu]

I've been using a locally modified version of SpikeInterface that exposes save_preprocessed_copy, but I would love to see support for this feature in an official release.

Does this PR have a chance of being merged (after the spelling is fixed)?