sweep.yaml

# This sweep file can be used to run hyper-parameter search using Weight & Biases tools
# See: https://docs.wandb.com/sweeps
program: reinforcement_learning/multi_agent_training.py
method: bayes
metric:
    name: evaluation/smoothed_score
    goal: maximize
parameters:
    n_episodes:
        values: [2000]
    hidden_size:
        # default: 256
        values: [128, 256, 512]
    buffer_size:
        # default: 50000
        values: [50000, 100000, 500000, 1000000]
    batch_size:
        # default: 32
        values: [16, 32, 64, 128]
    training_env_config:
        values: [0, 1, 2]