Create Table and helper functions to get position interval name from Interval list key and/or epoch

[samuelbray32]

Addresses issue of position interval not always being "pos {epoch-1} valid times".

• Introduces a new table PositionIntervalMap in common_behav composed of IntervalList primary keys and position_interval_name secondary key
• Updates the convert_epoch_interval_name_to_position_interval_name function from the decoding core to make use of this table and moves it to common_behav. Function populates PositionIntervalMap if map is not found, meaning shouldn't need to populate table elsewhere if you use this function.
• convert_epoch_interval_name_to_position_interval_name can find position interval using nwb_file_name and either epoch or `interval_list_name'
• Cleanup cases where position interval defined assuming epoch-1 with calls to this function

[edeno]

Seems good to me. I had some comments which are more just things to think about but I didn't see anything that I'm sure would fail.

[edeno]

So the assumption here is that the position time interval is always contained within the interval (with some error). I think we have some other code just checking for any overlap in the intervals (in common interval) which might be more robust?

[samuelbray32]

Yeah looking for values from interval_list_intersect would be an easy way to check if there's any overlap. I don't know if there's use cases where you would want it to return a value for partial overlap, or scenarios where there could be multiple small position intervals in an epoch though? This solution seemed the most strict but more likely to give 1:1 mapping

[edeno]

There can be gaps in the position intervals due to loss of tracking but I agree that it probably won't be a problem

[edeno]

Just to note a different way this has been done (via regex which I don't necessarily like):

spyglass/src/spyglass/common/common_interval.py

Lines 115 to 119 in 482a652

	pos_valid_times = (
	interval_list.set_index("interval_list_name")
	.filter(regex=r"^pos \d+ valid times$", axis=0)
	.valid_times
	).sort_index(key=lambda index: [int(name.split()[1]) for name in index])

[dpeg22]

Any chance you've tested this on Abhilasha's NWB file with only run epochs?

[dpeg22]

Is it worth having a dependence on TaskEpoch to have epoch as a primary key as well?

[samuelbray32]

Not sure if it makes any use cases trickier. It would require you to know the epoch at population, which may not be as obvious for derived interval_list_names (?). Epoch could also have a one:many relationship with interval_list_name, so you'd still need a way to select which to use when searching by epoch

[dpeg22]

You could potentially raise an Exception here instead