Views
No views yet
SoccerTwos utilizando Unity ML-Agents.SoccerTwos (archivo .zip descargable)1git clone --depth 1 https://github.com/Unity-Technologies/ml-agents
21cd ml-agents
2pip3 install -e ./ml-agents-envs
3pip3 install -e ./ml-agents1mkdir -p ./train-soccer
21import gdown
2file_id = '1KuqBKYiXiICU4kNMqEzhgyuPF5_45CL'
3gdown.download(f'https://drive.google.com/uc?export=download&id={file_id}', './train-soccer/SoccerTwos.zip', quiet=False) unzip -d ./train-soccer ./train-soccer/SoccerTwos.zip```bash
chmod -R 755 ./train-soccer/SoccerTwos/SoccerTwos.x86_64behaviors:
SoccerTwos:
trainer_type: poca
hyperparameters:
batch_size: 4096
buffer_size: 40960
learning_rate: 0.0002
beta: 0.01
epsilon: 0.15
lambd: 0.92
num_epoch: 5
learning_rate_schedule: linear
network_settings:
normalize: true
hidden_units: 1024
num_layers: 3
vis_encode_type: simple
reward_signals:
extrinsic:
gamma: 0.995
strength: 1.0
keep_checkpoints: 10
max_steps: 1000000
time_horizon: 1200
summary_freq: 5000
self_play:
save_steps: 50000
team_change: 300000
swap_steps: 3000
play_against_latest_model_ratio: 0.6
initial_elo: 1200.0with open('./train-soccer/SoccerTwos.yaml', 'w') as file:
file.write(yaml_content)mlagents-learn ./train-soccer/SoccerTwos.yaml --env=./train-soccer/SoccerTwos/SoccerTwos.x86_64 --run-id="SoccerTwosNew" --no-graphics --force