Refined version of ymliao98/PS_MPI
- Use config.yml to simplify settings
- Simplified process to make it general
- Cut out useless code to make it elegant
- add your own datasets and models
- create corresponding dataset_name.py in /datasets
- define it in /datasets/dataset_name.py and return train\test dataset transformed
- import it in /datasets/init.py
- create corresponding model_name.py in /models
- define it in /models/model_name.py and return model class
- import it in /models/init.py and decide when to use your model in create_model_instance()
The number after should be larger than config.yml:num_workers
mpiexec --oversubscribe -n 1 python server.py : -n 10 python client.py
This will put newest result in /results and also delete empty runs dirs. Don't use when training not finished.
python analysis_tools/generate_results.py
Modify this according to your need. The current version will print required results into on figure. Edit it as your wish. python analysis_tools/draw_pics.py
WARNING: Clear all logs!
rm -rf model_save server_log clients_log config_save pycache