Prep: weekly mtg 20160316 (Byron, Steve, me) - mobeets/nullSpaceControl GitHub Wiki

Days with saturating behavior

• plots of thresholds for all metrics
• mostly the same threshold times as Pete had...
• all Jeffy

dtsGood = {'20120709', '20120601', '20120331', '20120327', '20120308', '20120525'};

Dates were chosen based on there being asymptotes in either trial_length or cursor progress for either all or all but one targetAngle condition. Also, there must be a global asymptote. (Note that no sessions from 2013, monkey 2, met these conditions.)

Progress vs. trial length

...

Fits over time

Looking at hypothesis performance as a function of time.

Sometimes the cloud/habitual fits get better along with performance:

And sometimes they get worse:

Also, note that cloud-hab and usually habitual also are better than kinematics mean in predicting intuitive block activity. So the worse-and-worse activity is indicative of a different hypothesis becoming more true, while for sessions like 20120709 it's as if it's resuming being the best hypothesis.

So for 20131125, for example, this one also gets worse with time for cloud hab. BUT! If you instead do the rotated cloud hypothesis, this model stays the same over time, and is a better hyp.

Note how initially, no rotation is the best model, but with time, the -45 model becomes the best.

Neural metrics over time

• CCA
• proportion YR of YN+YR

Most look more like this: The norm(YR) increases, norm(YN) stays roughly the same, and the CCA increases slightly.

20120525 is kinda weird.