Using output from #10, perform a PCA to determine components that covary.
The first step is to raster-scan the affinity matrices from "video" format into a 2D matrix, with frames as columns and "affinities" as rows; if you had 100 GMM components and 100 frames over which you evaluated those components, your raster-scanned matrix would be 10,000 x 100.
Perform PCA on this matrix to reduce the time dimension and discover covariances between the components over time. Un-raster-scan the corresponding eigenvectors back into "affinity"-shaped matrices to see where the covariances are positive/negative, and use that implant that information over the original 2D spatial plots of GMM components.
Compare these covariances between conditions (more details TBD).
Using output from #10, perform a PCA to determine components that covary.
The first step is to raster-scan the affinity matrices from "video" format into a 2D matrix, with frames as columns and "affinities" as rows; if you had 100 GMM components and 100 frames over which you evaluated those components, your raster-scanned matrix would be 10,000 x 100.
Perform PCA on this matrix to reduce the time dimension and discover covariances between the components over time. Un-raster-scan the corresponding eigenvectors back into "affinity"-shaped matrices to see where the covariances are positive/negative, and use that implant that information over the original 2D spatial plots of GMM components.
Compare these covariances between conditions (more details TBD).