Chunking-as-Policy-Compression

Code for producing results in Chunking as Policy Compression project.

Questions? Contact [email protected].

Main Functios for Experiment 1 (Set size manipulation)

analyze_rawdata.m

First step of any analysis. It converts raw jsPsych experiment data saved in .csv files to MATLAB data structures. Usage:

data = analyze_rawdata('setsize_manip')

learning_curve.m

Plot the task performance against the training length.

time_on_task.m

Plot the average response time against the training length.

exploratory_analysis_exp1.m

Exploratory analysis and plotting on the average accuracy, average RT, intrachunk RT in different blocks of experiment 1. Usage:

exploratory_analysis_exp1(plotCase, data)

where plotCase is a string of the analysis to be conducted. It can be 'avgAcc', 'avgRT', or 'intrachunkRT'.

policy_complexity_analysis_exp1.m

Analyses related to policy complexity, including average policy complexity in different blocks, reward-complexity curves, rain cloud plot of policy complexity distribution in different blocks, and a bunch of statistical tests.

fit_models.m

Model fitting. Usage:

[results,bms_results] = fit_models(models, data)

where models is a cell array of names of the model variants to be fitted, including "no_cost". "no_cost_chunk", "fixed", "fixed_chunk", "adaptive", "adaptive_chunk".

sim_from_empirical.m

Simulate data using fitted model parameters of the best fitted model. Usage:

simdata = sim_from_empirical()

We can then inspect the behavior of the simulated data using exploratory_analysis_exp1() and learning_curve().

Main Functions for Experiment 2 (Load & Incentive manipulation)

analyze_rawdata.m

Converts raw jsPsych experiment data saved in .csv files to MATLAB data structures. Usage:

data = analyze_rawdata('modified_freq_discr')

Use the specifier 'modified_freq_discr' for the load & incentive manipulation experiment with modified frequency discrimination task.

exploratory_analysis_exp2.m

Exploratory analysis and plotting on the average accuracy, average RT, intrachunk RT in different blocks of experiment 2. Usage:

exploratory_analysis_exp1(plotCase, data)

where plotCase is a string of the analysis to be conducted. It can be 'avgAcc', 'avgRT', or 'intrachunkRT'.

policy_complexity_analysis_exp2.m

Average policy complexity and reward-complexity curves in different blocks.

sim_rc_tradeoff.m

Simulate and plot the average reward, the policy complexity, and the reward-complexity tradeoff under the load and incentive manipulation conditions.