training-workflows
Use AgileRL classical single-agent training workflows for Gymnasium PPO, DQN, RainbowDQN, DDPG, TD3, replay/rollout buffers, and distributed training setup.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- training-workflows
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use AgileRL classical single-agent training workflows for Gymnasium PPO, DQN, RainbowDQN, DDPG, TD3, replay/rollout buffers, and distributed training setup.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md2.2 KB
- 参考文档references/configuration.md1.7 KB
- 参考文档references/distributed-training.md1.2 KB
- 参考文档references/troubleshooting.md1.7 KB
- 参考文档references/workflows.md3.1 KB
03 SKILL.md 正文
AgileRL Training Workflows
Use this sub-skill for AgileRL classical single-agent RL training with Gymnasium environments: PPO on-policy, DQN/RainbowDQN/DDPG/TD3 off-policy, recurrent variants, population setup, replay or rollout buffers, and safe pre-flight checks before long runs.
Read First
references/workflows.mdfor on-policy/off-policy training recipes and population flow.references/api-reference.mdfor important training helper and buffer APIs.references/configuration.mdforINIT_HP,net_config, YAML, logging, and environment patterns.references/distributed-training.mdfor Accelerate/distributed constraints.references/troubleshooting.mdfor shape, dependency, vector-env, W&B, and runtime failures.scripts/agilerl_classic_smoke.py --helpfor a no-training import/config probe.
Route Within AgileRL
| Need | Route |
|---|---|
| Evolution/tournament/mutation configuration | ../hpo-and-mutation/SKILL.md |
Custom net_config, EvolvableNetwork, CNN/LSTM/MultiInput setup |
../evolvable-modules/SKILL.md |
| Multi-agent PettingZoo training | ../multi-agent-and-wrappers/SKILL.md |
| Offline datasets or contextual bandits | ../offline-bandits-data/SKILL.md |
| LLM fine-tuning | ../llm-fine-tuning/SKILL.md |
Standard Classical Training Flow
- Install AgileRL and verify imports; use the root
scripts/check_agilerl_install.pyif unsure. - Create or wrap a Gymnasium environment; prefer
make_vect_envs(env_name, num_envs=...)for vectorized classical training. - Define
INIT_HPwith algorithm-specific keys andPOP_SIZE. - Define
net_configwithencoder_configandhead_configunless the default network is sufficient. - Create a population with
create_population(algo=..., observation_space=..., action_space=..., net_config=..., INIT_HP=..., population_size=..., num_envs=..., device=...). - For off-policy algorithms, create
ReplayBuffer; for on-policy PPO, use rollout collection/training helpers rather than replay memory. - Configure
TournamentSelectionandMutationsin the HPO sub-skill. - Call
train_on_policy(...)ortrain_off_policy(...), or implement a custom loop followingreferences/workflows.md.
Algorithm Selection
- Use
PPOfor on-policy, stable policy-gradient workflows and recurrent PPO variants. - Use
DQNorRainbowDQNfor discrete-action off-policy value learning. - Use
DDPGorTD3for continuous-action off-policy actor-critic workflows. - Use
train_on_policyfor PPO andtrain_off_policyfor DQN/RainbowDQN/DDPG/TD3 families.
Safe Validation
Run a smoke check before training:
python scripts/agilerl_classic_smoke.py --algorithm DQN --env CartPole-v1
python scripts/agilerl_classic_smoke.py --algorithm PPO --env CartPole-v1
The helper checks imports, Gymnasium spaces, basic config shape, and algorithm family routing. It does not train or download anything.
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.