SSS: view geometrically

[mobeets]

visualize all quantities in ratio, along with ellipses during int. and pert. for each theta.

and view the two marginals as well: joint in dotted line, compared to theta-conditional for each hypothesis.

[mobeets]

The following is the cov(Yh*SSS) for each Yh compared to cov(Y*SSS) of observed activity during the perturbation, grouped by row space angles from intuitive session. Each row is a different session, each column a different θ.

[mobeets]

covErrors arranged just as above, where errors have same scale, and anything > 5 is yellow

[mobeets]

finally, taking the average across directions, and then the average across session, for each hyp. errorbars are inner 50th percentile across sessions.

[mobeets]

finally, here is the comparison between the observed cov in SSS when activity was relevant (during intuitive, in gray) and when it was irrelevant (during perturbation, in black):

[mobeets]

now log of the trace ratio for irrelevant/relevant: < 0 (more red) means it shrank when it became irrelevant, > 0 (more blue) means it expanded.

and the mean of these values per session:

[mobeets]

whereas cloud usually predicts that activity will expand:

how could this be? cloud just resamples intuitive activity, so if it ends up sampling extreme values of the cloud more often, this might result in a stretching of variance?

[mobeets]

[mobeets]

[mobeets]