train_x10.py

from datetime import datetime
from pathlib import Path

import utils
from DQN import DQNAgent

seed = 100
save_dir = Path("result") / f"DQN-{datetime.now().strftime(r'%Y-%m-%d-%H-%M-%S')}"
utils.start_tensorboard(save_dir)
utils.setup_seed(seed)

# 创建模型类
model = DQNAgent(
    "envs:GridWorld",
    env_size=10,
    seed=seed,
    hidden_dim=80,
    buffer_capicity=50000,
    buffer_init_ratio=0.999,
    total_steps=200000,
    lr=5e-2,
    lr_gamma=0.20,
    lr_miletones=[100000 // 4 * i for i in range(1, 4)],
    tau=0.005,
    update_interval=5,
    gamma=0.90,
    batch_size=512,
    save_dir=save_dir,
    optimal_q_table="result/optimal_qv_x10.npy",
)
# 启动训练
model.train()