Examples for how to train appear in /scripts
.
- Go into the scripts directory
cd scripts
- Generate some training and validation boards of size 5, for the "network" mode:
python generate.py --num-boards 10000 -n 5 -p 0.65 -o train_data_n5.npz --mode 0
python generate.py --num-boards 100 -n 5 -p 0.65 -o validation_data_n5.npz --mode 0
- Train a network on the GPU using the example hyper parameters specified
in
examples/config.yml
python train.py -c ../examples/config.yml -v validation_data_n5.npz -t train_data_n5.npz -o train_results_n5.pkl.gz --device cuda
To use different parameters, just modify the YAML file.