`restrict_to_interval` does not work for individual trials

[raeedcho]

Trying to use restrict_to_interval on an individual trial (which is a Pandas Series, rather than a DataFrame) doesn't seem to work (many of the functions used by restrict_to_interval require a DataFrame input).

In fact, it seems that perhaps many functions would fail if given a single trial (e.g. get_time_varying_fields).

A possible solution might be to put in explicit checks in functions for single trial input instead of a DataFrame. I'm not sure if that would need to be done in every function that assumes a DataFrame, or if there's a more elegant solution available.

[bagibence]

Yes, one of the reasons for this is that currently anything that creates a copy of the input will fail if the first positional argument or the "trial_data" keyword argument is not a DataFrame because the @utils.copy_td decorator throws an error in this case.

I guess turning the Series into a DataFrame with one row might fix some of these. The problem with that is when applying functions to a Series, you expect them to return a Series (and not a DataFrame), right?

[raeedcho]

Hmm, good point. I think the best solution would be to explicitly turn my individual trial from a Series into a one-row DataFrame before running functions on it meant for multi-trial data and leave it at that. I probably should have thought of this sooner, except I'm very new to working with pandas, so I don't know its ins and outs yet 😁

I'm happy to close this issue with that solution though, if you agree.

[bagibence]

Don't close it just yet :) It's a good point (and issue) and the code should work on Series objects as well.
I was thinking that probably the easiest way of solving this is that if the functions get a Series as the input, they convert it to a DataFrame, do their thing, then -- maybe only some of them -- convert it back to a Series (instead of returning a one-row DataFrame. I guess this might be done within utils.copy_td or with an additional decorator?

What I would suggest is that first we organize the code a bit and move everything that is now just dumped into utils and tools into smaller modules, then return to this. Doing that might make it a bit more clear which parts of the codebase should be modified to fix this. What do you think?

[raeedcho]

Splitting the code into smaller modules certainly makes sense to me, and I agree that maybe there should be a decorator that does the Series->DataFrame->Series wrapping conversion in the case that the input is only a single trial.