Decoding analysis

[alexpiet]

Goals

• Determine the extent to which neural activity is behaviorally relevant for each strategy

Option 1

• For each cell, build a dataframe with filtered events 0,750ms relative to image onset. For either change or pre-change image.
• Then use a random forest decoder to predict change vs non-change
• Increase number of cells used (np paper used 20)

Option 2

• Decode whether the mouse licked or not
• Is licking responses more correlated with v1 in visual?
• you could do this just on changes, so then its the same as decoding hits vs miss

[alexpiet]

https://github.com/corbennett/ephys_behavior/blob/master/analysis_Sam.py

[alexpiet]

• How do I subselect, or sample multiple combinations of cells within each session? Do I sample like ___ times, then for each sample the random forest classifer is sampling from again over image presentations?
• Then how do I aggregate scores?

[alexpiet]

• I think Sam/Corbett equally weight each sample of combinations of cells.