foraging2 translator is missing change times

[dougollerenshaw]

@neuromusic @mochic : No catch trials are being identified when loading Foraging2 data with the master branch. When I revert to the 'fix/load_licks' branch and reload the same PKL file, the problem resolves, so it would appear not to be an issue with the underlying Foraging2 data.

Minimum code to replicate error (on master branch):

from visual_behavior.translator.foraging2 import data_to_change_detection_core
from visual_behavior.change_detection.trials.extended import create_extended_dataframe
from visual_behavior.visualization.extended_trials.daily import make_daily_figure
import pandas as pd

datapath=r"\\allen\programs\braintv\workgroups\neuralcoding\Behavior\Data\M347745\output\180430100756650000.pkl"

data=pd.read_pickle(datapath)

core_data=data_to_change_detection_core(data)

trials = create_extended_dataframe(
    trials=core_data['trials'], 
    metadata=core_data['metadata'], 
    licks=core_data['licks'], 
    time=core_data['time'],
)

assert len(trials[trials['trial_type']=='catch'])>0

[neuromusic]

@dougollerenshaw there's only one line changed on the fix/lload_licks branch & it should not affect this issue

indeed, when I use the fix/lload_licks branch, I still see the error you note on the file you are using.

[neuromusic]

drilling into this file more, it looks like there simply weren't any non-aborted catch trials in this session...

no_change = ~pd.isnull(core_data['trials']['change_time'])
print (core_data['trials'][no_change]['initial_image_name']==core_data['trials'][no_change]['change_image_name']).sum()

yeilds 0

[dougollerenshaw]

Hmm...
We are talking about this file, right?:\allen\programs\braintv\workgroups\neuralcoding\Behavior\Data\M347745\output\180430100756650000.pkl

on 'fix/load_licks' branch:

len(trials[
    (trials['trial_type']=='catch')&
    (trials['trial_type']!='aborted')
])

returns 50

on master:

len(trials[
    (trials['trial_type']=='catch')&
    (trials['trial_type']!='aborted')
])

returns 0

[neuromusic]

I get 0 on fix/load_licks for those expressions.

are you using an older version of that branch?

[dougollerenshaw]

Yes, I was using an older version of that branch. Sorry. Now that you've merged it in, it's displaying the same incorrect behavior for me. But there are catch trials in this session. I just went back to the 'foraging-2' branch that Chris committed on 4/20/18 at 1:06 and that is giving me 50 catch trials.

[dougollerenshaw]

Well, it looks like I'm wrong. There don't appear to be any catch trials in this session. When I look directly at the core_data['trials'] object, I can't find any trials where the initial_image_name matches the final_image_name. This returns an empty dataframe:

core_data['trials'][
    ~pd.isnull(core_data['trials']['initial_image_name'])&
    (core_data['trials']['initial_image_name']==core_data['trials']['change_image_name'])
]

That would partially explain issue 65.

Tagging @derricw. Are we correct that there are no catch trials being generated by sessions from real mice on cluster D?

[neuromusic]

note: the core_data['trials'] object is not "direct"... it is the object we are building from derric's foraging2 output

[dougollerenshaw]

Is it possible that core_data['trials'] is missing the catch trials somehow?

[neuromusic]

yes, absolutely. I'm investigating to see where it might have broken

[neuromusic]

found the problem. working on a fix.

[dougollerenshaw]

OK. One more bit of information. In this particular PKL file, I looked in core_data['trials'] for all changes with a first lick and a change time. Trial index 37 looks like a catch trial, but we don't have access to the initial and change stimuli in order to confirm that.

core_data['trials'][
    ~pd.isnull(core_data['trials']['first_lick'])&
    ~pd.isnull(core_data['trials']['change_time'])
][['first_lick','change_time','initial_image_name','change_image_name']][:10]

Gives:

[dougollerenshaw]

Just had a quick conversation with @mochic about this. He's on it! Thanks Chris.