This repository contains the necessary code to replicate the results presented in my report 'Evaluating Inverse Contextual Bandits against model-based inference' based on an ICML 2022 paper 'Inverse Contextual Bandits: Learning How Behavior Evolves over Time'.
The original work of Alihan Hüyük et al. is available at this github repository.
My proposed modifiaction of the Bayesian ICB (B-ICB) for the optimistic and greedy policies are available in src/main-optimistic-bicb.py and src/main-greedy-bicb.py.
Main experimental results can be obtained by running
./run.sh
python src/main-mle-model-and-eval.py
python src/optim-greedy-bicb.py
Inference algorithms for the stationary, linear, stepping and regressing models together with their evaluation (part of Tables 1 and 2) are available in src/main-mle-model-and-eval.py
Ealuation of the original, optimistic and greedy versions of B-ICB (second part of Table 1 and 2, Figure 1, and Table 3) can be found in src/optim-greedy-bicb.py.
Note: In order to run the experiments access to semi-synthetic data as generated in the original paper is needed.