SPO - Single Pass Optimization

Single Pass Optimization creates MLP models for single-step target point generation for passing in Soccer Simulation 2D.

Requirements:

• conda
• download the dataset used in the paper.
• download the BPS model used in the paper and place it inside a models folder.

How to setup SPO:

    >> install and activate the environment
    conda env create -f ./environment.yaml
    conda activate spo

    >> unpack the dataset used in the paper
    unzip ./db.zip -d ./

Model configuration

To configure the model, modify the hparams.yaml file.

Train

Note: to reduce training time, the training is not memory-optimized. Thus, a system with higher memory may be required.

    >> to generate a new BPS model:
    python train_cyrus.py train

    >> run the SPO training
    python train.py