binsmoother

binsmoother(Responses, SigE, BackgroundProb, NumberSteps)

BINSMOOTHER EM learning-curve estimation for binary (correct/incorrect) trial data

Usage:

[p05, p95, pmid, pmode, pmatrix, xnew, signewsq] = …

binsmoother(Responses, SigE, BackgroundProb, NumberSteps)

Inputs:

Responses : 1xK double - number correct per trial (row or column) – required SigE : double - initial guess for state random-walk SD – required BackgroundProb : double - chance-level probability of correct response – required NumberSteps : integer - maximum number of EM iterations – required

Outputs:

p05 : 1xK+1 double - lower 5 percent confidence bound on p(correct) p95 : 1xK+1 double - upper 95 percent confidence bound on p(correct) pmid : 1xK+1 double - median p(correct) pmode : 1xK+1 double - mode of the p(correct) density pmatrix : K+1x1 double - certainty that performance exceeds chance per trial xnew : 1xK+1 double - smoothed latent-state estimate x{k|K} signewsq : 1xK+1 double - smoothed state variance SIG^2{k|K}

Notes

Runs forward filter -> backward smoother -> EM M-step until the change in the estimated process variance (and initial state, for UpdaterFlag >= 1) falls below 1e-8. UpdaterFlag is hard-coded to 2 (no prior chance bias). MaxResponse is inferred from max(Responses).

See also: forwardfilter, backwardfilter, em_bino, pdistn

∿∿∿ Prerau Laboratory MATLAB Codebase · sleepEEG.org ∿∿∿