how does filtering change results?

[mobeets]

turn off angError filtering and see how things change. is habitual still good?

remember, angError is huge in Blk2 and not that big in Blk1:

[mobeets]

okay, wow, turns out, if you stop filtering out places where the actual cursor kinematic doesn't match the goal kinematic (i.e., angle between cursor and target), the scores usually get better. especially volitional (for all four sessions). in one session (one of the 2013 ones) this actually makes volitional better than habitual!

with and without filtering on angularError:

[mobeets]

this is even true for (gasp) the 20120601 day, if you do the volitional using the first two factors!

[mobeets]

for 20120601, before and after filtering on angError:

you know what, this makes a lot of sense: the ones with big angular errors are more likely the ones where the monkey was using the wrong internal mapping, i.e., ones earlier in the perturbation perhaps.

[mobeets]

and for volitional:

[mobeets]

but it's not just about using times more near the perturbation:

[mobeets]

more smoothed, and now for both blocks:

[mobeets]

the definitive guide to how filtering affects our results:

note that "no-filter" = anyAngErr + noMin + noMax

[mobeets]

note that volitional is always best with anyAngErr + noMin, and in 3 out of 4 cases this makes it better than the habitual hypothesis (in terms of error -- probably not in terms of cov).