/Neural-Linear-Bandits-with-Likelihood-Matching

Primary LanguagePython

Online Limited Memory Neural-Linear Bandits with Likelihood Matching

alt text

This library corresponds to the Online Limited Memory Neural-Linear Bandits with Likelihood Matching paper, ICML 2021.

The code is based on the "Deep Bayesian Bandits Showdown: An Empirical Comparison of Bayesian Deep Networks for Thompson Sampling" github repository.

Datasets

All the datasets can be found in UCI Machine Learning Repository, and should be placed under contextual_bandtis/datasets folder.

How to run the code?

Run at terminal:

 python3 main.py

How to configure the experiment?

At the main.py, set the method into:

  1. neural-linear-lm (Our method)
  2. neural-linear (full memory NeuralTS)
  3. linear (LinearTS)
  4. neural-linear-ntk (NTK version of limited memory NeuralTS)

At dataset sepcify the wanted dataset (an unknown dataset will cause an error). For amazon dataset, LinearTS do not work.

Requirements

  • tensorflow-gpu 1.15
  • absl-py 0.11
  • scipy 1.5.4