- torch >= 0.4.0
- tqdm
- numpy
- tensorboardX
mkdir data/; cd data
- Download CIFAR-10 (Link) in
data/
tar -xvf cifar-10-python.tar.gz
cd ..; python preprocess_cifar.py
cd code; python main.py
- Track all progress using tensorboard:
tensorboard --logdir code/tensorboard_stuff --port 16006
http://localhost:16006
preprocess_cifar.py
makes bandit_dataset from the original cifar dataset- Edit the file to update hyper_parameters like
num_sample
- Edit the file to update hyper_parameters like
- All the hyper-parameters for the banditnet implementation can be set in the file
code/hyper_params.py
Paper: https://www.cs.cornell.edu/people/tj/publications/joachims_etal_18a.pdf
BibTeX:
@InProceedings{Joachims/etal/18a,
author = {T. Joachims and A. Swaminathan and M. de Rijke},
title = {Deep Learning with Logged Bandit Feedback},
booktitle = {International Conference on Learning Representations (ICLR)},
year = 2018}