search

mobeets / nullSpaceControl

1 stars 0 forks source link

SSS: hypothesis predictions for det(cov) vs. learning #266

Closed mobeets closed 8 years ago

mobeets commented 8 years ago

see #256 and just do that for the different hypotheses

mobeets commented 8 years ago

'observed' 'cloud' 'habitual' 'unconstrained' 'baseline' 'minimum'

notice how unconstrained is basically near 1. this is probably because we're just consider det(cov) unconditional on target or anything, and so unconstrained will have all of the null space activity being completely preserved. other hyps will have covs changing due to different proportions of row space values being selected, between the two sessions.

screen shot 2016-08-11 at 10 21 00 am

mobeets commented 8 years ago

screen shot 2016-08-11 at 10 25 21 am

mobeets commented 8 years ago

if you instead do the detcov ratio conditional on thetaActualGrps, now of course habitual will be roughly 1.

x-axis is true data, y-axis is the three hyps (baseline/minimal are invisible in this view because their detcov is gigantic)

screen shot 2016-08-11 at 10 40 21 am