Skip to content

Latest commit

 

History

History
40 lines (27 loc) · 1.62 KB

readme.md

File metadata and controls

40 lines (27 loc) · 1.62 KB
Raw

CoLink - UniFed - FederatedScope

This repo integrates FederatedScope, a federated learning framework, into UniFed as a CoLink protocol. You can follow the steps below to test it.

1. Clone the repo

git clone https://github.com/HenryHu-H/colink-unifed-federatedscope.git

2. Create an environment

cd colink-unifed-federatedscope
conda create -n colink-unifed-federatedscope python=3.9
conda activate colink-unifed-federatedscope

3. Install the package in an editable mode

pip install -e .

4. Test the protocol

  • The first step is to write a test configuration. You can look into ./test/configs/case_0.json for an example. Note that for the case you construct, it should mainly serve the purpose of correctness testing (e.g. 1~2 epochs with a small model is usually sufficient). In this way, we can reproduce the correctness testing from a single host.

    • To check the output for running certain cases, change the case string target_case = "test/configs/case_0.json" in test_all_config.py (note that this only works when you install with -e flag), then, in the root directory of the repo, run
python test/test_all_config.py
  • Note that the vertical FL of FederatedScope only supports the standalone mode. Please run the above command first to load the dataset and convert the config file, and then run manually with
python federatedscope/federatedscope/main.py --cfg federatedscope/federatedscope/contrib/configs/vertical.yaml
  • Please do not run pytest directly as it may trigger other test cases provided by the FederatedScope framework.