stable-baselines3

# Stable Baselines3

## Overview

Stable Baselines3 (SB3) is a PyTorch-based library providing reliable implementations of reinforcement learning algorithms. This skill provides comprehensive guidance for training RL agents, creating custom environments, implementing callbacks, and optimizing training workflows using SB3's unified API.

## Core Capabilities

### 1. Training RL Agents

**Basic Training Pattern:**

```python
import gymnasium as gym
from stable_baselines3 import PPO

# Create environment
env = gym.make("CartPole-v1")

# Initialize agent
model = PPO("MlpPolicy", env, verbose=1)

# Train the agent
model.learn(total_timesteps=10000)

# Save the model
model.save("ppo_cartpole")

# Load the model (without prior instantiation)
model = PPO.load("ppo_cartpole", env=env)
```

**Important Notes:**
- `total_timesteps` is a lower bound; actual training may exceed this due to batch collection
- Use `model.load()` as a static method, not on an existing instance
- The replay buffer is NOT saved with the model to save space

**Algorithm Selection:**
Use `references/algorithms.md` for detailed algorithm characteristics and selection guidance. Quick reference:
- **PPO/A2C**: General-purpose, supports all action space types, good for multiprocessing
- **SAC/TD3**: Continuous control, off-policy, sample-efficient
- **DQN**: Discrete actions, off-policy
- **HER**: Goal-conditioned tasks

See `scripts/train_rl_agent.py` for a complete training template with best practices.

### 2. Custom Environments

**Requirements:**
Custom environments must inherit from `gymnasium.Env` and implement:
- `__init__()`: Define action_space and observation_space
- `reset(seed, options)`: Return initial observation and info dict
- `step(action)`: Return observation, reward, terminated, truncated, info
- `render()`: Visualization (optional)
- `close()`: Cleanup resources

**Key Constraints:**
- Image observations must be `np.uint8` in range [0, 255]
- Use channel-first format when possible (channels, height, width)
- SB3 normalizes images automatically by dividing by 255
- Set `normalize_images=False` in policy_kwargs if pre-normalized
- SB3 does NOT support `Discrete` or `MultiDiscrete` spaces with `start!=0`

**Validation:**
```python
from stable_baselines3.common.env_checker import check_env

check_env(env, warn=True)
```

See `scripts/custom_env_template.py` for a complete custom environment template and `references/custom_environments.md` for comprehensive guidance.

### 3. Vectorized Environments

**Purpose:**
Vectorized environments run multiple environment instances in parallel, accelerating training and enabling certain wrappers (frame-stacking, normalization).

**Types:**
- **DummyVecEnv**: Sequential execution on current process (for lightweight environments)
- **SubprocVecEnv**: Parallel execution across processes (for compute-heavy environments)

**Quick Setup:**
```python
from stable_baselines3.common.env_util import make_vec_env

# Create 4 parallel environments
env = make_vec_env("CartPole-v1", n_envs=4, vec_env_cls=SubprocVecEnv)

model = PPO("MlpPolicy", env, verbose=1)
model.learn(total_timesteps=25000)
```

**Off-Policy Optimization:**
When using multiple environments with off-policy algorithms (SAC, TD3, DQN), set `gradient_steps=-1` to perform one gradient update per environment step, balancing wall-clock time and sample efficiency.

**API Differences:**
- `reset()` returns only observations (info available in `vec_env.reset_infos`)
- `step()` returns 4-tuple: `(obs, rewards, dones, infos)` not 5-tuple
- Environments auto-reset after episodes
- Terminal observations available via `infos[env_idx]["terminal_observation"]`

See `references/vectorized_envs.md` for detailed information on wrappers and advanced usage.

### 4. Callbacks for Monitoring and Control

**Purpose:**
Callbacks enable monitoring metrics, saving checkpoints, implementing early stopping, and custom training logic without modifying core algorithms.

**Common Callbacks:**
- **EvalCallback**: Evaluate periodically and save best model
- **CheckpointCallback**: Save model checkpoints at intervals
- **StopTrainingOnRewardThreshold**: Stop when target reward reached
- **ProgressBarCallback**: Display training progress with timing

**Custom Callback Structure:**
```python
from stable_baselines3.common.callbacks import BaseCallback

class CustomCallback(BaseCallback):
    def _on_training_start(self):
        # Called before first rollout
        pass

    def _on_step(self):
        # Called after each environment step
        # Return False to stop training
        return True

    def _on_rollout_end(self):
        # Called at end of rollout
        pass
```

**Available Attributes:**
- `self.model`: The RL algorithm instance
- `self.num_timesteps`: Total environment steps
- `self.training_env`: The training environment

**Chaining Callbacks:**
```python
from stable_baselines3.common.callbacks import CallbackList

callback = CallbackList([eval_callback, checkpoint_callback, custom_callback])
model.learn(total_timesteps=10000, callback=callback)
```

See `references/callbacks.md` for comprehensive callback documentation.

### 5. Model Persistence and Inspection

**Saving and Loading:**
```python
# Save model
model.save("model_name")

# Save normalization statistics (if using VecNormalize)
vec_env.save("vec_normalize.pkl")

# Load model
model = PPO.load("model_name", env=env)

# Load normalization statistics
vec_env = VecNormalize.load("vec_normalize.pkl", vec_env)
```

**Parameter Access:**
```python
# Get parameters
params = model.get_parameters()

# Set parameters
model.set_parameters(params)

# Access PyTorch state dict
state_dict = model.policy.state_dict()
```

### 6. Evaluation and Recording

**Evaluation:**
```python
from stable_baselines3.common.evaluation import evaluate_policy

mean_reward, std_reward = evaluate_policy(
    model,
    env,
    n_eval_episodes=10,
    deterministic=True
)
```

**Video Recording:**
```python
from stable_baselines3.common.vec_env import VecVideoRecorder

# Wrap environment with video recorder
env = VecVideoRecorder(
    env,
    "videos/",
    record_video_trigger=lambda x: x % 2000 == 0,
    video_length=200
)
```

See `scripts/evaluate_agent.py` for a complete evaluation and recording template.

### 7. Advanced Features

**Learning Rate Schedules:**
```python
def linear_schedule(initial_value):
    def func(progress_remaining):
        # progress_remaining goes from 1 to 0
        return progress_remaining * initial_value
    return func

model = PPO("MlpPolicy", env, learning_rate=linear_schedule(0.001))
```

**Multi-Input Policies (Dict Observations):**
```python
model = PPO("MultiInputPolicy", env, verbose=1)
```
Use when observations are dictionaries (e.g., combining images with sensor data).

**Hindsight Experience Replay:**
```python
from stable_baselines3 import SAC, HerReplayBuffer

model = SAC(
    "MultiInputPolicy",
    env,
    replay_buffer_class=HerReplayBuffer,
    replay_buffer_kwargs=dict(
        n_sampled_goal=4,
        goal_selection_strategy="future",
    ),
)
```

**TensorBoard Integration:**
```python
model = PPO("MlpPolicy", env, tensorboard_log="./tensorboard/")
model.learn(total_timesteps=10000)
```

## Workflow Guidance

**Starting a New RL Project:**

1. **Define the problem**: Identify observation space, action space, and reward structure
2. **Choose algorithm**: Use `references/algorithms.md` for selection guidance
3. **Create/adapt environment**: Use `scripts/custom_env_template.py` if needed
4. **Validate environment**: Always run `check_env()` before training
5. **Set up training**: Use `scripts/train_rl_agent.py` as starting template
6. **Add monitoring**: Implement callbacks for evaluation and checkpointing
7. **Optimize performance**: Consider vectorized environments for speed
8. **Evaluate and iterate**: Use `scripts/evaluate_agent.py` for assessment

**Common Issues:**

- **Memory errors**: Reduce `buffer_size` for off-policy algorithms or use fewer parallel environments
- **Slow training**: Consider SubprocVecEnv for parallel environments
- **Unstable training**: Try different algorithms, tune hyperparameters, or check reward scaling
- **Import errors**: Ensure `stable_baselines3` is installed: `uv pip install stable-baselines3[extra]`

## Resources

### scripts/
- `train_rl_agent.py`: Complete training script template with best practices
- `evaluate_agent.py`: Agent evaluation and video recording template
- `custom_env_template.py`: Custom Gym environment template

### references/
- `algorithms.md`: Detailed algorithm comparison and selection guide
- `custom_environments.md`: Comprehensive custom environment creation guide
- `callbacks.md`: Complete callback system reference
- `vectorized_envs.md`: Vectorized environment usage and wrappers

## Installation

```bash
# Basic installation
uv pip install stable-baselines3

# With extra dependencies (Tensorboard, etc.)
uv pip install stable-baselines3[extra]
```

{
  "format": "aiskill/v1",
  "skill": {
    "name": "stable-baselines3",
    "slug": "stable-baselines3",
    "version": "1.0.0",
    "description": "Production-ready reinforcement learning algorithms (PPO, SAC, DQN, TD3, DDPG, A2C) with scikit-learn-like API. Use for standard RL experiments, quick prototyping, and well-documented algorithm implementations. Best for single-agent RL with Gymnasium environments. For high-performance parallel training, multi-agent systems, or custom vectorized environments, use pufferlib instead.",
    "category": "Research",
    "tags": [
      "research-science",
      "agents",
      "api",
      "performance"
    ],
    "platforms": [
      "chatgpt",
      "claude-code",
      "cursor"
    ],
    "tested_environment": "Agent Skills convention (SKILL.md) — works with Claude Code, Cursor, ChatGPT/Codex and compatible agents",
    "license": "MIT LICENSE"
  },
  "prompts": {
    "system": "# Stable Baselines3\r\n\r\n## Overview\r\n\r\nStable Baselines3 (SB3) is a PyTorch-based library providing reliable implementations of reinforcement learning algorithms. This skill provides comprehensive guidance for training RL agents, creating custom environments, implementing callbacks, and optimizing training workflows using SB3's unified API.\r\n\r\n## Core Capabilities\r\n\r\n### 1. Training RL Agents\r\n\r\n**Basic Training Pattern:**\r\n\r\n```python\r\nimport gymnasium as gym\r\nfrom stable_baselines3 import PPO\r\n\r\n# Create environment\r\nenv = gym.make(\"CartPole-v1\")\r\n\r\n# Initialize agent\r\nmodel = PPO(\"MlpPolicy\", env, verbose=1)\r\n\r\n# Train the agent\r\nmodel.learn(total_timesteps=10000)\r\n\r\n# Save the model\r\nmodel.save(\"ppo_cartpole\")\r\n\r\n# Load the model (without prior instantiation)\r\nmodel = PPO.load(\"ppo_cartpole\", env=env)\r\n```\r\n\r\n**Important Notes:**\r\n- `total_timesteps` is a lower bound; actual training may exceed this due to batch collection\r\n- Use `model.load()` as a static method, not on an existing instance\r\n- The replay buffer is NOT saved with the model to save space\r\n\r\n**Algorithm Selection:**\r\nUse `references/algorithms.md` for detailed algorithm characteristics and selection guidance. Quick reference:\r\n- **PPO/A2C**: General-purpose, supports all action space types, good for multiprocessing\r\n- **SAC/TD3**: Continuous control, off-policy, sample-efficient\r\n- **DQN**: Discrete actions, off-policy\r\n- **HER**: Goal-conditioned tasks\r\n\r\nSee `scripts/train_rl_agent.py` for a complete training template with best practices.\r\n\r\n### 2. Custom Environments\r\n\r\n**Requirements:**\r\nCustom environments must inherit from `gymnasium.Env` and implement:\r\n- `__init__()`: Define action_space and observation_space\r\n- `reset(seed, options)`: Return initial observation and info dict\r\n- `step(action)`: Return observation, reward, terminated, truncated, info\r\n- `render()`: Visualization (optional)\r\n- `close()`: Cleanup resources\r\n\r\n**Key Constraints:**\r\n- Image observations must be `np.uint8` in range [0, 255]\r\n- Use channel-first format when possible (channels, height, width)\r\n- SB3 normalizes images automatically by dividing by 255\r\n- Set `normalize_images=False` in policy_kwargs if pre-normalized\r\n- SB3 does NOT support `Discrete` or `MultiDiscrete` spaces with `start!=0`\r\n\r\n**Validation:**\r\n```python\r\nfrom stable_baselines3.common.env_checker import check_env\r\n\r\ncheck_env(env, warn=True)\r\n```\r\n\r\nSee `scripts/custom_env_template.py` for a complete custom environment template and `references/custom_environments.md` for comprehensive guidance.\r\n\r\n### 3. Vectorized Environments\r\n\r\n**Purpose:**\r\nVectorized environments run multiple environment instances in parallel, accelerating training and enabling certain wrappers (frame-stacking, normalization).\r\n\r\n**Types:**\r\n- **DummyVecEnv**: Sequential execution on current process (for lightweight environments)\r\n- **SubprocVecEnv**: Parallel execution across processes (for compute-heavy environments)\r\n\r\n**Quick Setup:**\r\n```python\r\nfrom stable_baselines3.common.env_util import make_vec_env\r\n\r\n# Create 4 parallel environments\r\nenv = make_vec_env(\"CartPole-v1\", n_envs=4, vec_env_cls=SubprocVecEnv)\r\n\r\nmodel = PPO(\"MlpPolicy\", env, verbose=1)\r\nmodel.learn(total_timesteps=25000)\r\n```\r\n\r\n**Off-Policy Optimization:**\r\nWhen using multiple environments with off-policy algorithms (SAC, TD3, DQN), set `gradient_steps=-1` to perform one gradient update per environment step, balancing wall-clock time and sample efficiency.\r\n\r\n**API Differences:**\r\n- `reset()` returns only observations (info available in `vec_env.reset_infos`)\r\n- `step()` returns 4-tuple: `(obs, rewards, dones, infos)` not 5-tuple\r\n- Environments auto-reset after episodes\r\n- Terminal observations available via `infos[env_idx][\"terminal_observation\"]`\r\n\r\nSee `references/vectorized_envs.md` for detailed information on wrappers and advanced usage.\r\n\r\n### 4. Callbacks for Monitoring and Control\r\n\r\n**Purpose:**\r\nCallbacks enable monitoring metrics, saving checkpoints, implementing early stopping, and custom training logic without modifying core algorithms.\r\n\r\n**Common Callbacks:**\r\n- **EvalCallback**: Evaluate periodically and save best model\r\n- **CheckpointCallback**: Save model checkpoints at intervals\r\n- **StopTrainingOnRewardThreshold**: Stop when target reward reached\r\n- **ProgressBarCallback**: Display training progress with timing\r\n\r\n**Custom Callback Structure:**\r\n```python\r\nfrom stable_baselines3.common.callbacks import BaseCallback\r\n\r\nclass CustomCallback(BaseCallback):\r\n    def _on_training_start(self):\r\n        # Called before first rollout\r\n        pass\r\n\r\n    def _on_step(self):\r\n        # Called after each environment step\r\n        # Return False to stop training\r\n        return True\r\n\r\n    def _on_rollout_end(self):\r\n        # Called at end of rollout\r\n        pass\r\n```\r\n\r\n**Available Attributes:**\r\n- `self.model`: The RL algorithm instance\r\n- `self.num_timesteps`: Total environment steps\r\n- `self.training_env`: The training environment\r\n\r\n**Chaining Callbacks:**\r\n```python\r\nfrom stable_baselines3.common.callbacks import CallbackList\r\n\r\ncallback = CallbackList([eval_callback, checkpoint_callback, custom_callback])\r\nmodel.learn(total_timesteps=10000, callback=callback)\r\n```\r\n\r\nSee `references/callbacks.md` for comprehensive callback documentation.\r\n\r\n### 5. Model Persistence and Inspection\r\n\r\n**Saving and Loading:**\r\n```python\r\n# Save model\r\nmodel.save(\"model_name\")\r\n\r\n# Save normalization statistics (if using VecNormalize)\r\nvec_env.save(\"vec_normalize.pkl\")\r\n\r\n# Load model\r\nmodel = PPO.load(\"model_name\", env=env)\r\n\r\n# Load normalization statistics\r\nvec_env = VecNormalize.load(\"vec_normalize.pkl\", vec_env)\r\n```\r\n\r\n**Parameter Access:**\r\n```python\r\n# Get parameters\r\nparams = model.get_parameters()\r\n\r\n# Set parameters\r\nmodel.set_parameters(params)\r\n\r\n# Access PyTorch state dict\r\nstate_dict = model.policy.state_dict()\r\n```\r\n\r\n### 6. Evaluation and Recording\r\n\r\n**Evaluation:**\r\n```python\r\nfrom stable_baselines3.common.evaluation import evaluate_policy\r\n\r\nmean_reward, std_reward = evaluate_policy(\r\n    model,\r\n    env,\r\n    n_eval_episodes=10,\r\n    deterministic=True\r\n)\r\n```\r\n\r\n**Video Recording:**\r\n```python\r\nfrom stable_baselines3.common.vec_env import VecVideoRecorder\r\n\r\n# Wrap environment with video recorder\r\nenv = VecVideoRecorder(\r\n    env,\r\n    \"videos/\",\r\n    record_video_trigger=lambda x: x % 2000 == 0,\r\n    video_length=200\r\n)\r\n```\r\n\r\nSee `scripts/evaluate_agent.py` for a complete evaluation and recording template.\r\n\r\n### 7. Advanced Features\r\n\r\n**Learning Rate Schedules:**\r\n```python\r\ndef linear_schedule(initial_value):\r\n    def func(progress_remaining):\r\n        # progress_remaining goes from 1 to 0\r\n        return progress_remaining * initial_value\r\n    return func\r\n\r\nmodel = PPO(\"MlpPolicy\", env, learning_rate=linear_schedule(0.001))\r\n```\r\n\r\n**Multi-Input Policies (Dict Observations):**\r\n```python\r\nmodel = PPO(\"MultiInputPolicy\", env, verbose=1)\r\n```\r\nUse when observations are dictionaries (e.g., combining images with sensor data).\r\n\r\n**Hindsight Experience Replay:**\r\n```python\r\nfrom stable_baselines3 import SAC, HerReplayBuffer\r\n\r\nmodel = SAC(\r\n    \"MultiInputPolicy\",\r\n    env,\r\n    replay_buffer_class=HerReplayBuffer,\r\n    replay_buffer_kwargs=dict(\r\n        n_sampled_goal=4,\r\n        goal_selection_strategy=\"future\",\r\n    ),\r\n)\r\n```\r\n\r\n**TensorBoard Integration:**\r\n```python\r\nmodel = PPO(\"MlpPolicy\", env, tensorboard_log=\"./tensorboard/\")\r\nmodel.learn(total_timesteps=10000)\r\n```\r\n\r\n## Workflow Guidance\r\n\r\n**Starting a New RL Project:**\r\n\r\n1. **Define the problem**: Identify observation space, action space, and reward structure\r\n2. **Choose algorithm**: Use `references/algorithms.md` for selection guidance\r\n3. **Create/adapt environment**: Use `scripts/custom_env_template.py` if needed\r\n4. **Validate environment**: Always run `check_env()` before training\r\n5. **Set up training**: Use `scripts/train_rl_agent.py` as starting template\r\n6. **Add monitoring**: Implement callbacks for evaluation and checkpointing\r\n7. **Optimize performance**: Consider vectorized environments for speed\r\n8. **Evaluate and iterate**: Use `scripts/evaluate_agent.py` for assessment\r\n\r\n**Common Issues:**\r\n\r\n- **Memory errors**: Reduce `buffer_size` for off-policy algorithms or use fewer parallel environments\r\n- **Slow training**: Consider SubprocVecEnv for parallel environments\r\n- **Unstable training**: Try different algorithms, tune hyperparameters, or check reward scaling\r\n- **Import errors**: Ensure `stable_baselines3` is installed: `uv pip install stable-baselines3[extra]`\r\n\r\n## Resources\r\n\r\n### scripts/\r\n- `train_rl_agent.py`: Complete training script template with best practices\r\n- `evaluate_agent.py`: Agent evaluation and video recording template\r\n- `custom_env_template.py`: Custom Gym environment template\r\n\r\n### references/\r\n- `algorithms.md`: Detailed algorithm comparison and selection guide\r\n- `custom_environments.md`: Comprehensive custom environment creation guide\r\n- `callbacks.md`: Complete callback system reference\r\n- `vectorized_envs.md`: Vectorized environment usage and wrappers\r\n\r\n## Installation\r\n\r\n```bash\r\n# Basic installation\r\nuv pip install stable-baselines3\r\n\r\n# With extra dependencies (Tensorboard, etc.)\r\nuv pip install stable-baselines3[extra]\r\n```",
    "user_template": ""
  },
  "variables": [],
  "install": {
    "steps": [
      {
        "title": "Download & unzip",
        "detail": "Download the .zip package from SkillVault and unzip it. The stable-baselines3/ folder contains SKILL.md plus any companion scripts, references and assets."
      },
      {
        "title": "Copy into your agent skills directory",
        "detail": "Move the folder into your agent's skills directory — e.g. ~/.claude/skills/ (Claude Code), .codex/skills/, .cursor/skills/ or your project's .claude/skills/ folder — then restart or reload the session."
      },
      {
        "title": "Use it",
        "detail": "Describe the task in natural language; the agent loads the skill automatically when the request matches the skill description."
      }
    ]
  },
  "usage": {
    "steps": []
  },
  "metadata": {
    "author": "SkillVault",
    "source": "https://skillvault.top/skills/stable-baselines3",
    "exported_at": "2026-11-03T07:41:38.702Z"
  }
}

format: aiskill/v1
skill:
  name: stable-baselines3
  slug: stable-baselines3
  version: 1.0.0
  description: >-
    Production-ready reinforcement learning algorithms (PPO, SAC, DQN, TD3, DDPG, A2C) with
    scikit-learn-like API. Use for standard RL experiments, quick prototyping, and well-documented
    algorithm implementations. Best for single-agent RL with Gymnasium environments. For
    high-performance parallel training, multi-agent systems, or custom vectorized environments, use
    pufferlib instead.
  category: Research
  tags:
    - research-science
    - agents
    - api
    - performance
  platforms:
    - chatgpt
    - claude-code
    - cursor
  tested_environment: >-
    Agent Skills convention (SKILL.md) — works with Claude Code, Cursor, ChatGPT/Codex and
    compatible agents
  license: MIT LICENSE
prompts:
  system: "# Stable Baselines3\r\n\r\n## Overview\r\n\r\nStable Baselines3 (SB3) is a PyTorch-based library providing reliable implementations of reinforcement learning algorithms. This skill provides comprehensive guidance for training RL agents, creating custom environments, implementing callbacks, and optimizing training workflows using SB3's unified API.\r\n\r\n## Core Capabilities\r\n\r\n### 1. Training RL Agents\r\n\r\n**Basic Training Pattern:**\r\n\r\n```python\r\nimport gymnasium as gym\r\nfrom stable_baselines3 import PPO\r\n\r\n# Create environment\r\nenv = gym.make(\"CartPole-v1\")\r\n\r\n# Initialize agent\r\nmodel = PPO(\"MlpPolicy\", env, verbose=1)\r\n\r\n# Train the agent\r\nmodel.learn(total_timesteps=10000)\r\n\r\n# Save the model\r\nmodel.save(\"ppo_cartpole\")\r\n\r\n# Load the model (without prior instantiation)\r\nmodel = PPO.load(\"ppo_cartpole\", env=env)\r\n```\r\n\r\n**Important Notes:**\r\n- `total_timesteps` is a lower bound; actual training may exceed this due to batch collection\r\n- Use `model.load()` as a static method, not on an existing instance\r\n- The replay buffer is NOT saved with the model to save space\r\n\r\n**Algorithm Selection:**\r\nUse `references/algorithms.md` for detailed algorithm characteristics and selection guidance. Quick reference:\r\n- **PPO/A2C**: General-purpose, supports all action space types, good for multiprocessing\r\n- **SAC/TD3**: Continuous control, off-policy, sample-efficient\r\n- **DQN**: Discrete actions, off-policy\r\n- **HER**: Goal-conditioned tasks\r\n\r\nSee `scripts/train_rl_agent.py` for a complete training template with best practices.\r\n\r\n### 2. Custom Environments\r\n\r\n**Requirements:**\r\nCustom environments must inherit from `gymnasium.Env` and implement:\r\n- `__init__()`: Define action_space and observation_space\r\n- `reset(seed, options)`: Return initial observation and info dict\r\n- `step(action)`: Return observation, reward, terminated, truncated, info\r\n- `render()`: Visualization (optional)\r\n- `close()`: Cleanup resources\r\n\r\n**Key Constraints:**\r\n- Image observations must be `np.uint8` in range [0, 255]\r\n- Use channel-first format when possible (channels, height, width)\r\n- SB3 normalizes images automatically by dividing by 255\r\n- Set `normalize_images=False` in policy_kwargs if pre-normalized\r\n- SB3 does NOT support `Discrete` or `MultiDiscrete` spaces with `start!=0`\r\n\r\n**Validation:**\r\n```python\r\nfrom stable_baselines3.common.env_checker import check_env\r\n\r\ncheck_env(env, warn=True)\r\n```\r\n\r\nSee `scripts/custom_env_template.py` for a complete custom environment template and `references/custom_environments.md` for comprehensive guidance.\r\n\r\n### 3. Vectorized Environments\r\n\r\n**Purpose:**\r\nVectorized environments run multiple environment instances in parallel, accelerating training and enabling certain wrappers (frame-stacking, normalization).\r\n\r\n**Types:**\r\n- **DummyVecEnv**: Sequential execution on current process (for lightweight environments)\r\n- **SubprocVecEnv**: Parallel execution across processes (for compute-heavy environments)\r\n\r\n**Quick Setup:**\r\n```python\r\nfrom stable_baselines3.common.env_util import make_vec_env\r\n\r\n# Create 4 parallel environments\r\nenv = make_vec_env(\"CartPole-v1\", n_envs=4, vec_env_cls=SubprocVecEnv)\r\n\r\nmodel = PPO(\"MlpPolicy\", env, verbose=1)\r\nmodel.learn(total_timesteps=25000)\r\n```\r\n\r\n**Off-Policy Optimization:**\r\nWhen using multiple environments with off-policy algorithms (SAC, TD3, DQN), set `gradient_steps=-1` to perform one gradient update per environment step, balancing wall-clock time and sample efficiency.\r\n\r\n**API Differences:**\r\n- `reset()` returns only observations (info available in `vec_env.reset_infos`)\r\n- `step()` returns 4-tuple: `(obs, rewards, dones, infos)` not 5-tuple\r\n- Environments auto-reset after episodes\r\n- Terminal observations available via `infos[env_idx][\"terminal_observation\"]`\r\n\r\nSee `references/vectorized_envs.md` for detailed information on wrappers and advanced usage.\r\n\r\n### 4. Callbacks for Monitoring and Control\r\n\r\n**Purpose:**\r\nCallbacks enable monitoring metrics, saving checkpoints, implementing early stopping, and custom training logic without modifying core algorithms.\r\n\r\n**Common Callbacks:**\r\n- **EvalCallback**: Evaluate periodically and save best model\r\n- **CheckpointCallback**: Save model checkpoints at intervals\r\n- **StopTrainingOnRewardThreshold**: Stop when target reward reached\r\n- **ProgressBarCallback**: Display training progress with timing\r\n\r\n**Custom Callback Structure:**\r\n```python\r\nfrom stable_baselines3.common.callbacks import BaseCallback\r\n\r\nclass CustomCallback(BaseCallback):\r\n    def _on_training_start(self):\r\n        # Called before first rollout\r\n        pass\r\n\r\n    def _on_step(self):\r\n        # Called after each environment step\r\n        # Return False to stop training\r\n        return True\r\n\r\n    def _on_rollout_end(self):\r\n        # Called at end of rollout\r\n        pass\r\n```\r\n\r\n**Available Attributes:**\r\n- `self.model`: The RL algorithm instance\r\n- `self.num_timesteps`: Total environment steps\r\n- `self.training_env`: The training environment\r\n\r\n**Chaining Callbacks:**\r\n```python\r\nfrom stable_baselines3.common.callbacks import CallbackList\r\n\r\ncallback = CallbackList([eval_callback, checkpoint_callback, custom_callback])\r\nmodel.learn(total_timesteps=10000, callback=callback)\r\n```\r\n\r\nSee `references/callbacks.md` for comprehensive callback documentation.\r\n\r\n### 5. Model Persistence and Inspection\r\n\r\n**Saving and Loading:**\r\n```python\r\n# Save model\r\nmodel.save(\"model_name\")\r\n\r\n# Save normalization statistics (if using VecNormalize)\r\nvec_env.save(\"vec_normalize.pkl\")\r\n\r\n# Load model\r\nmodel = PPO.load(\"model_name\", env=env)\r\n\r\n# Load normalization statistics\r\nvec_env = VecNormalize.load(\"vec_normalize.pkl\", vec_env)\r\n```\r\n\r\n**Parameter Access:**\r\n```python\r\n# Get parameters\r\nparams = model.get_parameters()\r\n\r\n# Set parameters\r\nmodel.set_parameters(params)\r\n\r\n# Access PyTorch state dict\r\nstate_dict = model.policy.state_dict()\r\n```\r\n\r\n### 6. Evaluation and Recording\r\n\r\n**Evaluation:**\r\n```python\r\nfrom stable_baselines3.common.evaluation import evaluate_policy\r\n\r\nmean_reward, std_reward = evaluate_policy(\r\n    model,\r\n    env,\r\n    n_eval_episodes=10,\r\n    deterministic=True\r\n)\r\n```\r\n\r\n**Video Recording:**\r\n```python\r\nfrom stable_baselines3.common.vec_env import VecVideoRecorder\r\n\r\n# Wrap environment with video recorder\r\nenv = VecVideoRecorder(\r\n    env,\r\n    \"videos/\",\r\n    record_video_trigger=lambda x: x % 2000 == 0,\r\n    video_length=200\r\n)\r\n```\r\n\r\nSee `scripts/evaluate_agent.py` for a complete evaluation and recording template.\r\n\r\n### 7. Advanced Features\r\n\r\n**Learning Rate Schedules:**\r\n```python\r\ndef linear_schedule(initial_value):\r\n    def func(progress_remaining):\r\n        # progress_remaining goes from 1 to 0\r\n        return progress_remaining * initial_value\r\n    return func\r\n\r\nmodel = PPO(\"MlpPolicy\", env, learning_rate=linear_schedule(0.001))\r\n```\r\n\r\n**Multi-Input Policies (Dict Observations):**\r\n```python\r\nmodel = PPO(\"MultiInputPolicy\", env, verbose=1)\r\n```\r\nUse when observations are dictionaries (e.g., combining images with sensor data).\r\n\r\n**Hindsight Experience Replay:**\r\n```python\r\nfrom stable_baselines3 import SAC, HerReplayBuffer\r\n\r\nmodel = SAC(\r\n    \"MultiInputPolicy\",\r\n    env,\r\n    replay_buffer_class=HerReplayBuffer,\r\n    replay_buffer_kwargs=dict(\r\n        n_sampled_goal=4,\r\n        goal_selection_strategy=\"future\",\r\n    ),\r\n)\r\n```\r\n\r\n**TensorBoard Integration:**\r\n```python\r\nmodel = PPO(\"MlpPolicy\", env, tensorboard_log=\"./tensorboard/\")\r\nmodel.learn(total_timesteps=10000)\r\n```\r\n\r\n## Workflow Guidance\r\n\r\n**Starting a New RL Project:**\r\n\r\n1. **Define the problem**: Identify observation space, action space, and reward structure\r\n2. **Choose algorithm**: Use `references/algorithms.md` for selection guidance\r\n3. **Create/adapt environment**: Use `scripts/custom_env_template.py` if needed\r\n4. **Validate environment**: Always run `check_env()` before training\r\n5. **Set up training**: Use `scripts/train_rl_agent.py` as starting template\r\n6. **Add monitoring**: Implement callbacks for evaluation and checkpointing\r\n7. **Optimize performance**: Consider vectorized environments for speed\r\n8. **Evaluate and iterate**: Use `scripts/evaluate_agent.py` for assessment\r\n\r\n**Common Issues:**\r\n\r\n- **Memory errors**: Reduce `buffer_size` for off-policy algorithms or use fewer parallel environments\r\n- **Slow training**: Consider SubprocVecEnv for parallel environments\r\n- **Unstable training**: Try different algorithms, tune hyperparameters, or check reward scaling\r\n- **Import errors**: Ensure `stable_baselines3` is installed: `uv pip install stable-baselines3[extra]`\r\n\r\n## Resources\r\n\r\n### scripts/\r\n- `train_rl_agent.py`: Complete training script template with best practices\r\n- `evaluate_agent.py`: Agent evaluation and video recording template\r\n- `custom_env_template.py`: Custom Gym environment template\r\n\r\n### references/\r\n- `algorithms.md`: Detailed algorithm comparison and selection guide\r\n- `custom_environments.md`: Comprehensive custom environment creation guide\r\n- `callbacks.md`: Complete callback system reference\r\n- `vectorized_envs.md`: Vectorized environment usage and wrappers\r\n\r\n## Installation\r\n\r\n```bash\r\n# Basic installation\r\nuv pip install stable-baselines3\r\n\r\n# With extra dependencies (Tensorboard, etc.)\r\nuv pip install stable-baselines3[extra]\r\n```"
  user_template: ""
variables: []
install:
  steps:
    - title: Download & unzip
      detail: >-
        Download the .zip package from SkillVault and unzip it. The stable-baselines3/ folder
        contains SKILL.md plus any companion scripts, references and assets.
    - title: Copy into your agent skills directory
      detail: >-
        Move the folder into your agent's skills directory — e.g. ~/.claude/skills/ (Claude Code),
        .codex/skills/, .cursor/skills/ or your project's .claude/skills/ folder — then restart or
        reload the session.
    - title: Use it
      detail: >-
        Describe the task in natural language; the agent loads the skill automatically when the
        request matches the skill description.
usage:
  steps: []
metadata:
  author: SkillVault
  source: https://skillvault.top/skills/stable-baselines3
  exported_at: "2026-11-03T07:41:38.702Z"