Export_to_ibl (take 2)

[jonahpearl]

Attempts to resolve #2843, again :) This version doesn't require any utility functions outside of what spikeinterface already has, plus some functionality from numpy and scipy.

I expect that someone who knows SI a bit better than me will be able to take the _get_rms function and convert it into a class that can use the ChunkExtractor to speed things up with multi-processing. If no one gets around to it by the next month I can have a go, I just have some deadlines in the meantime. The LFP spectrum is currently just calculated on a single snippet of LFP data, so I don't think it needs to be parallelized.

[zm711]

I think this is cool. I think Alessio would have to take the rms function to deal with the chunking. But keeping things within the si framework is super nice. a couple initial comments that should be super easy to do.

[alejoe91]

Will do! Thanks for the re-writing @jonahpearl !!

[jonahpearl]

-- Added a test (passes locally for me), made a few fixes based on that
-- Loosened the IBL unit standards a bit, since it's presumably ok to have some mua here. Could allow users to pass their own criteria?
-- Fixed skip_pc_metrics to be True (which was my original intention, just got the sign of the parameter flipped in my head). This should speed things up for large recordings, and if anyone really wants the PC metrics they can use the export to phy feature with those on.
-- Slightly simplified the bit that checks the pre-computed qm's.

[zm711]

Hey Jonah few more things on my end. I don't want to generate too much work before Alessio hops onto this PR. But I think we definitely need to protect for the case where the user no longer has a useable recording.

[zm711]

Thanks Jonah. I'll leave this alone now until Alessio deals with it just to ensure we don't end up with conflicting stuff! :)

[alejoe91]

Hi @jonahpearl

I finally found some time to review and work on this.

Here are some updates:

• I added parallelization to the compute rms function
• improved chunk retrieval for psd estimation
• remove export to phy and file mapping (I realized we have everything we need in the analyzer!)
• removed auto-computation of extensions and QC
• made the definition of good units more flexible, with a good_unit_query that acts on the qm dataframe
• extended tests
• renamed the function: export_to_ibl_gui (IMO, export to IBL is too vague)

Can you test it on your side and make sure that the IBL GUI works well with the output?

[jonahpearl]

Just tried it out — works well and fast. Seems like there's one test failing just because it still has the old function name.

[alejoe91]

Just tried it out — works well and fast. Seems like there's one test failing just because it still has the old function name.

Fixed now!

[alejoe91]

Awesome! I'll merge then :)

[jonahpearl]

Thanks Alessio!