Algorithm Distillation & Multi-Armed Bandits

This repository contents source code used to train & evaluate Transformer with Algorithm Distillation in a multi-armed bandit task.

Local installation

First of all, make sure that you have python>=3.10installed.

cd $HOME && git clone https://github.com/minashkinvladislav/ad.git
cd $HOME/ad
python3.10 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Usage

To run experiments with pretrained model:

cd ./experiments
python run.py --mode=load

To train model by yourself and then run experiments:

cd ./experiments
python run.py --mode=train

Custom model will be saved at saved_models/custom.

Results

Results can be found in figures folder .