Deep reinforcement learning in a racket sport for player evaluation with technical and tactical contexts

Author

Ning Ding

Reference:

Ning Ding, Kazuya Takeda, Keisuke Fujii, Deep reinforcement learning in a racket sport for player evaluation with technical and tactical contexts, IEEE Access, accepted.

Requirements:

• Python 2.7

• Numpy 1.16.5

• Tensorflow 1.14.0

• Scipy

• Matplotlib

• scikit-learn

External Dependencies:

• AlphaPose
• TrackNet

Data:

• Video data can be downloaded from Youtube. To run our model directly, you can also download the preprocessd data from Dropbox

Usage

Training:

1. Modify the save_mother_dir in configuration.py as your save directory
2. Cd into your save_mother_dir, make two directories ./models/hybrid_sl_saved_NN/ and ./models/hybrid_sl_log_NN/
3. Download the preprocessd data.
4. Run python Train.py
5. The trained model will be saved in the file (e.g. ./saved_models_gammaXX_hdXX_iterXX_lrXX)

Evaluation:

1. Run python Evaluate.py
2. To obtain the result of action value in a badminton rally. Run python plot.py --iter_number xx

Acknowledgements:

For this project, we relied on research codes from:

• MVIG-SJTU/AlphaPose
• TrackNet
• DRL-ice-hockey