尝试不同的控制环境

本次运行结果

Dependency cache ready: /Users/sanbu/Documents/Codex/2026-10-20/revie/work/chapter-lab-runs/studio-zh/workspaces/gymnasium/.cache/online-experiments/hands-on-modern-rl-experiment-gymnasium-75bfe84ee278.ready
代码仓库： /Users/sanbu/Documents/Codex/2026-10-20/revie/work/chapter-lab-runs/studio-zh/workspaces/gymnasium
实验运行目录： /Users/sanbu/Documents/Codex/2026-10-20/revie/work/chapter-lab-runs/studio-zh/workspaces/gymnasium/modelscope-space/hands-on-modern-rl-experiment-gymnasium

可选实验：
  Bandit · ε-greedy
  Blackjack · Monte Carlo
  GridWorld · Q-Learning
  FrozenLake · Q-Learning
  CliffWalking · SARSA
  Taxi · Q-Learning
  CartPole · DQN
  CartPole · PPO
  MountainCar · Tabular Q
  Acrobot · PPO
  Pendulum · PPO
  MountainCarContinuous · SAC
轮次安排：2 轮 × 1,000 steps

      实时训练日志
      Multi-Armed Bandit training console
========================================================================
    0.0s  CONFIG  steps=2000 epochs=2 alpha=0.1 epsilon=0.1 seed=42
    0.3s  EPOCH   1/2 step=1000/2000 average_reward=0.665 best_arm=3
    0.3s  SAVE    model_id=1792631041-003000-epoch-01

      实时训练日志
      Multi-Armed Bandit training console
========================================================================
    0.0s  CONFIG  steps=2000 epochs=2 alpha=0.1 epsilon=0.1 seed=42
    0.3s  EPOCH   1/2 step=1000/2000 average_reward=0.665 best_arm=3
    0.3s  SAVE    model_id=1792631041-003000-epoch-01
    0.3s  EPOCH   2/2 step=2000/2000 average_reward=0.723 best_arm=3
    0.3s  SAVE    model_id=1792631041-003000-epoch-02

      实时训练日志
      Multi-Armed Bandit training console
========================================================================
    0.0s  CONFIG  steps=2000 epochs=2 alpha=0.1 epsilon=0.1 seed=42
    0.3s  EPOCH   1/2 step=1000/2000 average_reward=0.665 best_arm=3
    0.3s  SAVE    model_id=1792631041-003000-epoch-01
    0.3s  EPOCH   2/2 step=2000/2000 average_reward=0.723 best_arm=3
    0.3s  SAVE    model_id=1792631041-003000-epoch-02

<Figure size 820x400 with 1 Axes>

结果文件： /Users/sanbu/Documents/Codex/2026-10-20/revie/work/chapter-lab-runs/studio-zh/workspaces/gymnasium/modelscope-space/hands-on-modern-rl-experiment-gymnasium/artifacts/bandit-run-summary.json
已保存 {len(playground.load_models(EXPERIMENT))} 个此任务可选择的轮次模型。

VBox(children=(Dropdown(description='训练轮次模型：', layout=Layout(width='95%'), options=(('轮次 2/2 · 2,000/2,000 步 ·…

默认选择最新轮次；也可以选择其他已保存模型，然后点击 回放所选策略.

{
  "score": 0.723,
  "steps": 2000,
  "epochs": 2,
  "checkpoints": 2,
  "preview": "learned-policy.png"
}

{
  "score": 0.723,
  "steps": 2000,
  "epochs": 2,
  "checkpoints": 2,
  "preview": "learned-policy.png"
}