Paper revisions collaboration with Marina

[yavorska-iryna]

Iryna's main priorities:

• area and depth analysis for all cre lines

most of the code that is used for area and depth analysis is in GLM repo, cluster module

• selecting number of clusters

I tried doing F-test, jupyter notebook attached. The results of this test were interpretable but the optimal N was slightly off from what we have seen with testing different N. We found that 12 clusters produced consistent interpretable results while F-test suggested 13+ clusters which leads to very small clusters (<10 cells).

Marina has worked on the tasks below:

• shuffle controls for all cre clusters
• clustering on folds of GLM
• familiar only vs novel clusters

[yavorska-iryna]

PR that included optimal K cluster testing : https://github.com/AllenInstitute/visual_behavior_analysis/pull/841