Using output from #8, compute a "video" of affinity matrices for each cell trajectory.
This means: using the method for computing an affinity matrix that Andrew used in the SciPy paper (evaluating each GMM component according to the other component's PDF; we could also experiment with KL-divergence here if we want), build an affinity matrix at that frame/time point. Build out the affinity matrices at all the frames over which the GMM components were evaluated and "stack" the matrices together like a video.
There is a scaled version of the affinity table which will run on lists of means and covars so once the intermediates are created and dispersed this should be immediately doable.
Using output from #8, compute a "video" of affinity matrices for each cell trajectory.
This means: using the method for computing an affinity matrix that Andrew used in the SciPy paper (evaluating each GMM component according to the other component's PDF; we could also experiment with KL-divergence here if we want), build an affinity matrix at that frame/time point. Build out the affinity matrices at all the frames over which the GMM components were evaluated and "stack" the matrices together like a video.