Probabilistic formulation of the Take The Best heuristic
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data
results
README.md
discount_N.m
discount_independent.m
discount_info.m
flipnoise_sign_likelihood.m
plot_cue_rank_post_probs_and_validities.m
props_to_discrimination.m
props_to_pairwise_differences.m
r_algs.R
r_algs.m
run_linear_function_learning_experiment.m
run_tests_on_real_data.m
test_on_real_data.m
ttb_log_marg_lik.m
ttbfit.m
ttbmcmc.m
ttbmcmcpred.m
ttbmcmcremovewarmup.m
ttbpred.m

README.md

Probabilistic formulation of the Take The Best heuristic

This repository contains Matlab code implementing the probabilistic model of Take The Best heuristic and experiments described in

  • Peltola, Jokinen, Kaski. Probabilistic formulation of the Take The Best heuristic, to appear in the proceedings of CogSci2018.

Running the experiments

  • run_tests_on_real_data.m: accuracy comparisons on benchmark datasets (this can take a lot of time with 1000 repetitions; make n_reps smaller in test_on_real_data.m for faster results).
  • run_linear_function_learning_experiment.m: function learning task given biased (TTB generated) pairwise feedback.

The functions ttbfit (exact inference using exhaustive computation) and ttbmcmc (MCMC inference) can be used to learn the probabilistic TTB model from training data.

Requirements

The implementation of the probabilistic TTB does not have any requirements beyond Matlab.

Running the experiments requires:

Contact

Tomi Peltola, tomi.peltola@aalto.fi