domains / science / stable-baselines3-rl-training / stable-baselines3

stable-baselines3

Skill card

combined

The K-Dense copy is the more current, version-pinned procedure (SB3 2.9.0, Python 3.10+, PyTorch >= 2.8, Gymnasium 1.3.0, zsh quoting, MuJoCo extra, SB3-Contrib/SBX/RL-Zoo pointers, device/seed guidance, and a note that the custom-env template shape changed from (2,) to (4,)). The Vibe-Skills copy adds a routing boundary and a pufferlib alternative, and it contains a vectorized-env snippet that explicitly uses SubprocVecEnv, which the K-Dense copy does not show. Because each copy has material steps or constraints the other lacks, crowning one would hide a real alternative.

Repositories

Other copies

Combined skill

---
name: stable-baselines3
description: Trains and evaluates single-agent reinforcement learning with Stable Baselines3 (PPO, SAC, DQN, TD3, DDPG, A2C), Gymnasium custom environments, vectorized rollouts, callbacks, and checkpoint normalization. Applies to reproducible RL experiments, continuous control, discrete actions, and SB3-Contrib recurrent or masked policies. Use for standard RL experiments, quick prototyping, and well-documented algorithm implementations. For high-performance parallel training, multi-agent systems, or custom vectorized environments, use pufferlib instead.
license: MIT license
allowed-tools: Read Write Edit Bash
compatibility: Requires Python 3.10+, PyTorch >= 2.8, and stable-baselines3 2.9.0. Gymnasium environments; optional extras for TensorBoard and Atari (ale-py).
metadata:
  version: "2.0"
  last-reviewed: "2026-10-01"
  upstream-version: "2.9.0"
  skill-author: K-Dense Inc.
---

# Stable Baselines3

## Routing Boundary

Use this skill only for Stable-Baselines3, SB3, PPO/SAC/DQN, reinforcement learning agents, Gymnasium environments, policies, rollouts, and RL training workflows. Do not use it for ordinary scikit-learn, random forests, supervised classification, tabular regression, or generic machine-learning model training.

## 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.

**Current upstream:** SB3 **2.9.0** (June 15, 2026). Docs: [stable-baselines3.readthedocs.io](https://stable-baselines3.readthedocs.io/en/v2.9.0/).

## Installation

Tested against **stable-baselines3 2.9.0**. Requires **Python 3.10+** (3.9 dropped in 2.8.0) and **PyTorch >= 2.8**.

```bash
# Basic installation
uv pip install "stable-baselines3==2.9.0"

# With extra dependencies (TensorBoard, ale-py for Atari, etc.)
uv pip install "stable-baselines3[extra]==2.9.0"
```

The 2.9.0 release supports Gymnasium >=0.29.1,<2.0; this review exercised Gymnasium 1.3.0, PyTorch 2.14.1 and Python 3.13 on CPU. pandas/matplotlib are now optional extras. Installation requires network unless packages are cached; local RL training needs no credentials or service endpoints.

On zsh, quote brackets: `uv pip install 'stable-baselines3[extra]==2.9.0'`.

For MuJoCo continuous-control benchmarks:

```bash
uv pip install "gymnasium[mujoco]"
```

Check your version:

```python
import stable_baselines3
print(stable_baselines3.__version__)
```

## Related Projects

- **[SB3-Contrib](https://github.com/Stable-Baselines-Team/stable-baselines3-contrib)**: experimental algorithms (MaskablePPO, CrossQ, QR-DQN, RecurrentPPO) — separate `sb3-contrib` package
- **[RL Baselines3 Zoo](https://github.com/DLR-RM/rl-baselines3-zoo)**: pre-trained agents, hyperparameters, training scripts
- **[SBX](https://github.com/araffin/sbx)**: SB3 + JAX implementations for users who prefer JAX over PyTorch

## 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 (device="cpu" is often faster for MlpPolicy on small envs)
model = PPO("MlpPolicy", env, verbose=1, device="cpu", seed=0)

# 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, device="cpu")
env.close()
```

**Important Notes:**
- `total_timesteps` is a lower bound; actual training may exceed this due to batch collection
- Call the class method `PPO.load(...)` and keep the returned new model
- 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 Box, Discrete, flat MultiDiscrete and MultiBinary actions, good for multiprocessing
- **SAC/TD3**: Continuous control, off-policy, sample-efficient
- **DQN**: Discrete actions, off-policy
- **HER**: Replay-buffer strategy for goal-conditioned off-policy 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:**
- Default CNN image preprocessing expects `np.uint8` in range [0, 255]
- Use channel-first format when possible (channels, height, width)
- SB3 normalizes images automatically by dividing by 255
- For pre-normalized float images, use channel-first layout and `policy_kwargs={"normalize_images": False}`
- 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 [the template](scripts/custom_env_template.py) and [environment guide](references/custom_environments.md).
The template now observes both agent and random goal coordinates, shape `(4,)`;
old shape `(2,)` checkpoints require retraining. `gym.make("CustomEnv-v0")` adds
the 100-step time limit; direct `CustomEnv()` does not. `check_env` checks API
consistency, not Markov sufficiency, reward correctness or learnability.

### 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 import PPO
from stable_baselines3.common.env_util import make_vec_env

# DummyVecEnv batches 4 lightweight environments sequentially.
env = make_vec_env("CartPole-v1", n_envs=4, seed=0)
try:
    model = PPO("MlpPolicy", env, verbose=1, device="cpu")
    model.learn(total_timesteps=25000)
finally:
    env.close()
```

For compute-heavy environments, use SubprocVecEnv:

```python
from stable_baselines3.common.vec_env import SubprocVecEnv

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.

## Conflicts

- K-Dense-AI/scientific-agent-skills::skills/stable-baselines3/SKILL.md pins stable-baselines3 2.9.0, Python 3.10+, PyTorch >= 2.8, and Gymnasium >=0.29.1,<2.0, and includes zsh bracket quoting and MuJoCo installation; foryourhealth111-pixel/Vibe-Skills::bundled/skills/stable-baselines3/SKILL.md has no version pin and no installation section.
- K-Dense-AI/scientific-agent-skills::skills/stable-baselines3/SKILL.md says PPO/A2C support Box, Discrete, flat MultiDiscrete and MultiBinary actions; foryourhealth111-pixel/Vibe-Skills::bundled/skills/stable-baselines3/SKILL.md says PPO/A2C support all action space types.
- K-Dense-AI/scientific-agent-skills::skills/stable-baselines3/SKILL.md shows make_vec_env with n_envs=4 and seed=0 without specifying vec_env_cls; foryourhealth111-pixel/Vibe-Skills::bundled/skills/stable-baselines3/SKILL.md shows make_vec_env with vec_env_cls=SubprocVecEnv.
- K-Dense-AI/scientific-agent-skills::skills/stable-baselines3/SKILL.md includes a routing note that the custom-env template now observes both agent and random goal coordinates with shape (4,) and that old (2,) checkpoints require retraining; foryourhealth111-pixel/Vibe-Skills::bundled/skills/stable-baselines3/SKILL.md does not mention this shape change.