0xcryptohuang / ee-net-iclr-2022

EE-Net: Exploitation-Exploration Neural Networks in Contextual Bandits

In this repository, we provide one implementation of EE-Net, where the linear decision-maker (e.g., f1 + f2 ) and neural decision-maker (f3) are provided respectively. For the exploration network, one-layer CNN is used to reduce the dimensionality of gradient of exploitation network.

Run:

Run EE-Net on Mnist:

python EENet_run.py

Run baselines on Mnist:

python baselines/baselines_run.py

Prerequisites:

python 3.8.8, CUDA 11.2, torch 1.9.0, torchvision 0.10.0, sklearn 0.24.1, numpy 1.20.1, scipy 1.6.2, pandas 1.2.4

Hyper-parameters

dim: dimensionality of arm context vector

n_arm: number of arms.

pooling_step_size: aggregation size for the gradient, and the aggregated gradient will be the input of f2

hidden: width of all neural networks