To run the bandit algorithm:

python3 run.py --config_path config/toy/mala.json --exp_idx 0

This command reads the configuration you want to evaluate (check in folder config/toy), generates you synthetic data. And then It runs the experiment and save the result in the folder result. The attribute idx_exp allows to run your experiment many times on several devices using a Slurm array without overwriting your result file.

Firstly, we have to load the Yahoo dataset: "R6A - Yahoo! Front Page Today Module User Click Log Dataset, version 1.0 (1.1 GB)". The dataset can be found in https://webscope.sandbox.yahoo.com/catalog.php?datatype=r.

Then, to run the yahoo experiment:

python3 run.py --config_path config/yahoo/fgmala_yahoo.json --exp_idx 0

This command is similar to the one for the toy example.

This experiment is based on the folder Logistic. Firstly, you have to generate the synthetic data. Please, go on the folder Logistic and run

python3 data_generator.py --config configs/gauss_bandit_baseline.yaml

Then, run bandit algorithm using the command

python3 run_simulation.py --config_path configs/logistic/lmc.yaml --repeat 10 --log