ShuttleNet: Position-aware Rally Progress and Player Styles Fusion for Stroke Forecasting in Badminton (AAAI 2022)

Official code of the paper ShuttleNet: Position-aware Rally Progress and Player Styles Fusion for Stroke Forecasting in Badminton. Paper: https://arxiv.org/abs/2112.01044

Update

This repo has been moved to CoachAI Projects and will no longer be maintained.

Overview

ShuttleNet contains two encoder-decoder modified Transformer as extractors, and a position-aware gated fusion network for fusing these contexts to tackle stroke forecasting in badminton.

Setup

• Build environment

  conda env create -f environment.yml
  

• Put data in data/dataset.csv

Run script

• To run all the cases, run the script script_all.sh:

  ./script_all.sh {model_type} {sample_num}
  

• To run the case with different seeds, run the script script_seed.sh:

  ./script_all.sh {model_type} {encode_length} {sample_num}
  

• To run the case with the specific seed and default samples, run the script script_single.sh:

  ./script_single.sh {model_type} {encode_length} {seed_value}
  

• You may need to modify model_type in train.py choices.

Output files

• Trained models will be saved in the folder model/{datetime}_{model_type}/
  • Details of performance will also be contained in folder named result.log
• The performance will be saved in corresponding log file named {model_type}_result_{encode_length}

Code structure

Training

python train.py {model_type}

Evaluate

python evaluate.py {model_path}

Generate

python generator.py {model_path}

Citation

If you use our dataset or find our work is relevant to your research, please cite:

@misc{ShuttleNet_2022_AAAI,
    author = {Wei-Yao Wang, Hong-Han Shuai, Kai-Shiang Chang, Wen-Chih Peng},
    title = {ShuttleNet: Position-aware Fusion of Rally Progress and Player Styles for Stroke Forecasting in Badminton},
    year = {2021}
    eprint = {2112.01044},
    archivePrefix = {arXiv},
    primaryClass = {cs.CL}
}