This repository contains code realted to SCEPTIC, which was developed to model how humans explore and learn a complex reinforcement-based timing task. For more information see Dombrovski Hallquist Selective Maintenance and Entropy-Driven Exploration Facilitate Human Learning of Temporal Instrumental Contingencies.
Please note Matlab is required to run SCEPTIC
You must have the VBA tool box downloaded and installed, for instructions see here for more information see here
Clone or download this repo
1.) Create a config file by running:
s=create_scecptic_configuration_struct
2a.) Execute the parent function to run model(s) on all subjects
sceptic_fit_group_vba(sceptic_config_file)
2b.) Conversely run model(s) on a single subject
[posterior,out] = clock_sceptic_vba(s,id,model,data_file)
3.) Compare models
sceptic_grp_BMCFixed learning rate, value-based choice
selective maintenance, fixed learning rate, value-based choice, action values decay
fixed learning rate, choice controlled by weighted sum of value and uncertainty
Kalam filter (KF) learning rule, value-based choice
KF learning rule, choice controlled by weighted sum of value and uncertainty
Which version of the clock task is being analyzed Default: hallquist_clock
Path pointing to processed subjects' data Format: csv
Path pointing to where to save indivdual results outputted from VBA toolbox Outputs: posterior out
Path pointing to where to save the group results
Cell containing strings of model names Default: {'fixed' 'fixed_decay' 'fixed_uv' 'kalman_softmax' 'kalman_uv_sum'}
Number of basis functions Default: 24
VBA option defning the likelihood function of what you are tyring to predict Default: 1
VBA option if you would like to vary/keep constant parameters during muliple ecxperimental sessions Default: 0
If when using multisession you would like to keep all model parameters fixed across multiple runs Default 0
If you would like the toolbox to fit the prop_spread parameter Default: 0
Number of time bins Default:50
Allow for uncertainty aversion in UV_sum Default: 1
If you would like to save the results Default: 0
If you would like to display graphics 0
Take std over all timesteps and possible draws per condition for sigma_noise parameter Default: []
Maximum time point of reaction time range (on bin scale (ex 4000ms in 10's time bin scale is 400)) Default: 400
| Alex Dombrovski | Michael Hallquist | Jonathan Wilson |