mixedlearningcurve2¶
- mixedlearningcurve2(N, Z, background_prob, rhog, alphag, betag, sig2eg, sig2vg, startflag, binflag)
MIXEDLEARNINGCURVE2 EM learning-curve estimation with configurable initial condition
- Usage:
- [alph, beta, gamma, rho, sig2e, sig2v, xnew, signewsq, muone, a] = …
- mixedlearningcurve2(N, Z, background_prob, rhog, alphag, betag, …
sig2eg, sig2vg, startflag, binflag)
- Inputs:
N : 1xK double - observations at each trial – required Z : 1xK double - continuous observation (e.g. reaction time) – required background_prob : double - chance-level probability of correct response – required rhog : double - initial guess for state AR(1) coefficient rho – required alphag : double - initial guess for RT intercept alpha – required betag : double - initial guess for RT slope beta – required sig2eg : double - initial guess for RT noise variance sig2e – required sig2vg : double - initial guess for state variance sig2v – required startflag : integer - initial-condition rule: 0 fixes xnew(1)=0 (no bias),
2 sets xnew(1)=xnew(2) (no prior chance probability), other values leave the smoother initial condition unchanged – required
binflag : logical - if true, suppress RT model (force alpha=beta=0) – required
- Outputs:
alph : double - EM estimate of RT intercept beta : double - EM estimate of RT slope gamma : double - binary observation state weight (fixed to 0 here) rho : double - EM estimate of state AR(1) coefficient sig2e : double - EM estimate of RT noise variance sig2v : double - EM estimate of state variance xnew : 1xK+1 double - smoothed state estimate x{k|K} signewsq : 1xK+1 double - smoothed state variance SIG^2{k|K} muone : double - logit of background probability a : 1xK+1 double - smoother gain A{k}
Notes
Variant of mixedlearningcurve with selectable initial-condition handling (startflag) and a binary-only mode (binflag). Iterates up to 3000 EM steps with convergence criterion 1e-6 on mean absolute change in [alpha, beta, sig2e, sig2v]. Originally authored by Anne Smith (Oct 15, 2003); updated by Anne Smith (Nov 29, 2010) and Michael Prerau.
See also: mixedlearningcurve, recfilter, backest, m_step
∿∿∿ Prerau Laboratory MATLAB Codebase · sleepEEG.org ∿∿∿