Logistic regression

Julia code for fitting various logistic regression models to the rat behavioral data from the Poisson clicks task. Allows us to assess L/R bias, as well as temporal discounting. Uses the GLM.jl package.

In short, fits the following four models:

1D click difference

2D, total number of right and left clicks

Click difference across time

Number of right and left clicks across time

Producing, for each rat, a figure that looks like the following:

To do

• Major refactor to make the code easier to use
• Use stratified (across gammas) kfold cv for the estimation of params and their error bars
  • Make sure that the X matrices contain gammas
• Check that other generalization metrics also show same trends as AUC(ROC)
• Visualize how separable the w * X + beta distributions are prior to feeding into the logit model
• Metarat analysis: could fit to one giant collated matrix or average logit weights
• Explore different bin sizes - are 20Hz rats fit worse bc of low click count per bin?