search

interpreting-rl-behavior / interpreting-rl-behavior.github.io

Code for the site https://interpreting-rl-behavior.github.io/
Creative Commons Attribution 4.0 International
0 stars 0 forks source link

Plot temporal cross correlation matrices for PC dims #17

Closed leesharkey closed 2 years ago

leesharkey commented 2 years ago

Even though we've used PC to reduce the dimensionality of the space that we want to interpret, it's still too many dimensions for our puny human brains to notice patterns in easily.

One way to help find patterns that we want to look is to identify patterns in the data automatically and then we can go study them in more detail.

A good way to do this is to plot cross correlations between PCs at different timesteps.

We'd plot a cross correlation matrix between PC activations at t=T and PC activations at t=T+K (for K in {-k, ... , -1, 0, 1, ... , k}. Mostly this will just be weak noise, but any correlations that are strong is suggestive of some causal relationship between them.

Later if we're feeling ambitious we can extend this to include PCs in the environment-subspace.

N.B. The code for this will be in the other repo, not this website repo. But I'm putting this issue here so we keep all our issues in one place.

leesharkey commented 2 years ago

Resolved by commit 8b5b33baddc268e9c65a61e3d237456ab48722e3 in the other repo. Keeping issues from the other repo on the website repo in the interest of having all project issues in one place.