vectorization
Create, step, wrap, and troubleshoot Gymnasium vectorized environments for batched training and evaluation.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- vectorization
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Create, step, wrap, and troubleshoot Gymnasium vectorized environments for batched training and evaluation.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/troubleshooting.md6.5 KB
- 参考文档references/vector-envs.md7.1 KB
- 参考文档references/vector-wrappers-and-spaces.md6.5 KB
03 SKILL.md 正文
Gymnasium Vectorization
Use this sub-skill when a task needs multiple Gymnasium environments stepped as one batch: gymnasium.make_vec, SyncVectorEnv, AsyncVectorEnv, vector wrappers, batched observations/actions/rewards/masks/infos, autoreset modes, or vector multiprocessing troubleshooting.
Start Here
- Prefer
gymnasium.make_vec(env_id, num_envs=..., vectorization_mode="sync" | "async")for registered environments. - Use
envs.action_space.sample()or another action batch matchingenvs.action_space; useenvs.single_action_spaceonly to reason about one sub-environment. - Treat
rewards,terminations, andtruncationsas arrays of shape(num_envs,), not scalar booleans. - Check
envs.metadata["autoreset_mode"]before interpreting post-terminal observations or writing bootstrapping logic. - Use
gymnasium.wrappers.vectorwrappers around vector envs; pass single-env wrappers throughmake_vec(..., wrappers=[...])only when they should wrap each sub-environment before vectorization. - Always call
envs.close(), especially forAsyncVectorEnvworker processes.
Route by Task
| Task | Read |
|---|---|
| Create or step vector envs, choose sync vs async, handle batched outputs | references/vector-envs.md |
| Apply vector wrappers, transform batched spaces, use vector space utilities | references/vector-wrappers-and-spaces.md |
| Debug actions, masks, autoreset, multiprocessing, shared memory, render/video | references/troubleshooting.md |
| Prove a small local vector loop works | scripts/vector_env_smoke.py |
Minimum Correct Pattern
import gymnasium as gym
envs = gym.make_vec("CartPole-v1", num_envs=4, vectorization_mode="sync")
try:
observations, infos = envs.reset(seed=123)
actions = envs.action_space.sample()
observations, rewards, terminations, truncations, infos = envs.step(actions)
done_mask = terminations | truncations
finally:
envs.close()
For a quick sanity check from this sub-skill directory, run python scripts/vector_env_smoke.py --help and then python scripts/vector_env_smoke.py --mode sync --num-envs 2 --steps 4.
Scope Boundaries
- Use
../environment-api/SKILL.mdfor singleEnvcontracts, custom environment classes, registration,check_env, and non-vector reset/step migration. - Use
../spaces-data/SKILL.mdfor fundamentalSpaceconstructors, dtype/shape design,contains, flattening, and JSONable conversion. - Use
../wrappers-recording/SKILL.mdfor single-environment wrappers, recording, rendering, and custom wrapper subclassing depth. - Use
../builtin-envs/SKILL.mdfor built-in family selection, optional extras/backends, Atari/ALE, Box2D, MuJoCo, and Toy Text action masks.
Keep runtime guidance self-contained. Do not depend on the Gymnasium source tree, tests, or documentation paths at runtime.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.