GWeindel / hmp

Repository for the hmp python package
BSD 3-Clause "New" or "Revised" License
35 stars 8 forks source link

Dev #132

Closed GWeindel closed 8 months ago

GWeindel commented 8 months ago

Simplified fit function and removed deltas

I removed all the parameters that weren't used just to simplify, this PR removed

More importantly, based on simulations it looks like the delta parameter used up to now is only necessary for the detection of the first event (as there is a bias in the likelihood when we split the RTs in two). It doesn't seem necessary to have this counterfactual model for following events

removed filtering of non-converged models

as it seems addressed by recent progress in location definition and was leading to useless rejection of a good proposal by the fit function

Switched default location to event_width_samples

Overall choosing this default seems the best, fastest, strategy, if shorter stage duration are expected then this can be decreased or location_correlation_threshold can be used

Fixed problem with ultimate location not being 0

By forcing it to 0 in the propose_fit_params() function instead of the EM

minor changes

in README, tutorials and simulation functions