Skip to content
Gossip Algorithms and Decentralized SGD with Communication Compression
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
code
LICENSE
README.md

README.md

CHOCO-SGD

Code for the main experiments of the paper Decentralized Stochastic Optimization and Gossip Algorithms with Compressed Communication.

Datasets and Setup

First you need to download datasets from LIBSVM library and convert them into pickle format. For that from

cd data
wget -t inf https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/epsilon_normalized.bz2
wget -t inf https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/rcv1_test.binary.bz2
cd ../code
python pickle_datasets.py

If you get memory error, you can leave rcv1 dataset in the sparse format, but this would slow down training time.

Reproduce the results

For running experiments with the epsilon dataset

python experiment_epsilon_final.py final

Reference

If you use this code, please cite the following paper:

@inproceedings{ksj2019choco,
  author = {Anastasia Koloskova and Sebastian U. Stich and Martin Jaggi},
  title = {Decentralized Stochastic Optimization and Gossip Algorithms with Compressed Communication},
  booktitle = {ICML 2019 - Proceedings of the 36th International Conference on Machine Learning},
  url = {http://proceedings.mlr.press/v97/koloskova19a.html},
  series = {Proceedings of Machine Learning Research},
  publisher = {PMLR}, 
  volume = {97},
  pages = {3479--3487},
  year = {2019}
}
You can’t perform that action at this time.