ohhorange / PS_MPI

PS_MPI

1. Changes

Refined version of ymliao98/PS_MPI

1. Use config.yml to simplify settings
2. Simplified process to make it general
3. Cut out useless code to make it elegant
4. add your own datasets and models

2. Customize

1. Own dataset

1. create corresponding dataset_name.py in /datasets
2. define it in /datasets/dataset_name.py and return train\test dataset transformed
3. import it in /datasets/init.py

2. Own model

1. create corresponding model_name.py in /models
2. define it in /models/model_name.py and return model class
3. import it in /models/init.py and decide when to use your model in create_model_instance()

3. Run

The number after should be larger than config.yml:num_workers
mpiexec --oversubscribe -n 1 python server.py : -n 10 python client.py

4. Analysis tools

1. Save last training record to anther direction

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

2. Generate pictures

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

4. Clear ALL log and saved models

WARNING: Clear all logs!
rm -rf model_save server_log clients_log config_save pycache