events.load_trials doesn't work with stim length > 1 min

[MarvinT]

Lauren and @nvahidi have both run in to this issue.

couple possible solutions:

1. set default to 2 minutes
2. make it a parameter of events.load_trials
3. remove stim window altogether

any recommendations?

@neuromusic @zekearneodo @sasen @theilmbh

[theilmbh]

Either 2 or 3

[theilmbh]

Do 2. I think there will be a performance penalty in loading the trials for long recordings if it has to figure out for itself when stimuli might end. Unless you know a cleverer way to do 3.

[neuromusic]

here is roughly how to do option 3:

the trial parsing algorithm finds digmarks that indicate stim ends (> & #) a few lines earlier...
https://github.com/gentnerlab/ephys-analysis/blob/master/ephys/events.py#L275

I did this because it's more reliable than the stimstart digmark (<) b/c Spike2 sometimes drops multiple stim start digmarks ¯_(ツ)_/¯

so the trials dataframe is built from the digmark dataframe in the next line

so the info is grabbed, then discarded, then this get_stim_end is applied, which searches for the stimend digmarks.

who is the idiot who wrote this code!?

a better approach might be something like...

stim_end_mask = digmarks['codes'].isin(('>','#'))

trials = digmarks[stim_end_mask]['codes'].rename({'codes':'stimulus_end'})
# get_stim_start will need to be rewritten to return... something else. just the stim start time I guess?
trials['stimulus_start'] = trials.apply(lambda row: get_stim_start(row,digmarks),axis=1)
trials.reset_index(inplace=True)

# then do the rest of the trial info parsing...
trials['stimulus'] = trials.apply(lambda row: get_stim_info(row,stimulus,fs)['text'],axis=1)
trials['response'] = trials.apply(lambda row: get_response(row,digmarks,fs)['codes'],axis=1)
trials['response_time'] = trials.apply(lambda row: get_response(row,digmarks,fs)['time_samples'],axis=1)
trials['consequence'] = trials.apply(lambda row: get_consequence(row,digmarks,fs)['codes'],axis=1)
trials['correct'] = trials['consequence'].apply(is_correct)

[MarvinT]

I solved this using rigid_pandas. Not sure if load_trials ever got fixed. I'm not sure if people still use load_trials...