Summary: weekly mtg 20160518 (Byron, Steve, Matt, Aaron, me) - mobeets/nullSpaceControl GitHub Wiki

Byron: use unconditioned null space range, and conditioned range

  • do NOT use factor range because there's no biological interpretation (arbitrary rotations are just as viable)
    • (n.b. unconditioned range is already implicitly implemented, so just have conditioned range)
  • Q: does the factor range change the means? or just the variance?
  • to do: just include the tighter conditioned range--does that keep things like unconstrained reasonable hyps? or did we need the factor box fo that?

use thetaGrps or thetaActualGrps?

  • thetaActualGrps

ime cross-validated errors

  • not necessary for perturbation
  • for intuitive: fit on half, test on half
  • or, do the same after learning for remaining trials in perturbation

use finer range of thetaGrps

  • compare pruning and habitual in histograms

go from blk2 to blk1

  • pruning should not work going from blk2 to blk1
  • habitual should be worse too because sampling from too big a set

histograms of all hypotheses compared to observed at once

  • and on the bottom a little whisker plot of mean and one se on either side, for each hypothesis stacked

question: % in null for first 7 time points

  • larger than % in null after 7 time points?

future-direction: decoding kinematics activity in null space

  • for undergrad