evolvable-modules
Use AgileRL evolvable modules, networks, architecture configs, custom network wrappers, and mutation-compatible model building blocks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- evolvable-modules
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use AgileRL evolvable modules, networks, architecture configs, custom network wrappers, and mutation-compatible model building blocks.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md1.5 KB
- 参考文档references/configuration.md1.4 KB
- 参考文档references/modules-and-networks.md2.3 KB
- 参考文档references/troubleshooting.md1.5 KB
03 SKILL.md 正文
AgileRL Evolvable Modules
Use this sub-skill when a task asks for AgileRL EvolvableModule, EvolvableNetwork, net_config, MLP/CNN/LSTM/MultiInput/SimBa configuration, Q/value/actor networks, custom architectures, or wrapping ordinary PyTorch modules for AgileRL mutation compatibility.
Read First
references/modules-and-networks.mdfor conceptual relationships and build patterns.references/api-reference.mdfor key classes and config objects.references/configuration.mdforencoder_config,head_config, image/recurrent/multi-input settings.references/troubleshooting.mdfor shape, mutation, and custom module errors.scripts/inspect_evolvable_builders.py --helpfor safe tiny builder checks.
Boundaries
- Use
../training-workflows/SKILL.mdfor full Gymnasium training loops. - Use
../hpo-and-mutation/SKILL.mdfor mutation probabilities and tournament selection. - Use
../multi-agent-and-wrappers/SKILL.mdfor grouped multi-agentnet_configand PettingZoo agent IDs. - This sub-skill owns architecture selection, config validation, custom wrappers, and network construction.
Common Routes
| Task | Guidance |
|---|---|
| Vector observation to discrete action values | Use QNetwork with MLP encoder/head config. |
| Image observation policy/value network | Use CNN encoder config and verify channels/order. |
| Dict/Tuple observations | Use MultiInput config with MLP/CNN/LSTM sub-configs as needed. |
| Recurrent/partially observable workflow | Use LSTM config and read training recurrent notes. |
| Custom PyTorch module | Inherit EvolvableModule when architecture should mutate, or use DummyEvolvable when only RL hyperparameters/weights should mutate. |
| SimBa/ResNet style architecture | Use the corresponding module/network configs and validate dimensions before training. |
Safe Validation
python scripts/inspect_evolvable_builders.py --mode mlp
python scripts/inspect_evolvable_builders.py --mode dict
The helper constructs tiny Gymnasium spaces and config objects. It does not train.
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.