offline-bandits-data
Use AgileRL offline RL datasets, replay/data conversion, CQL/ILQL, contextual bandits, NeuralUCB/NeuralTS, and BanditEnv workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- offline-bandits-data
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use AgileRL offline RL datasets, replay/data conversion, CQL/ILQL, contextual bandits, NeuralUCB/NeuralTS, and BanditEnv workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md1.0 KB
- 参考文档references/data-formats.md1.5 KB
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.8 KB
- 脚本scripts/inspect_bandit_setup.py2.0 KB
03 SKILL.md 正文
AgileRL Offline RL And Bandit Data
Use this sub-skill for AgileRL offline reinforcement learning, replay/data conversion, CQL, ILQL, train_offline, contextual bandits, NeuralUCB, NeuralTS, and BanditEnv workflows.
Read First
references/workflows.mdfor offline RL and contextual bandit recipes.references/data-formats.mdfor transition, HDF5/Minari, and tabular bandit schemas.references/api-reference.mdfor replay/bandit/offline APIs.references/troubleshooting.mdfor shape, schema, and network/download failures.scripts/inspect_bandit_setup.py --helpfor a no-download synthetic bandit probe.
Boundaries
- Use
../training-workflows/SKILL.mdfor online Gymnasium PPO/DQN/DDPG/TD3 workflows. - Use
../hpo-and-mutation/SKILL.mdfor tournament/mutation configuration. - Use
../llm-fine-tuning/SKILL.mdfor language/LLM datasets and RL post-training environments. - This sub-skill owns static transition datasets, replay filling, Minari/HDF5 handling, and one-step contextual bandit datasets.
Offline RL Flow
- Load or validate a dataset with observations, actions, rewards, next observations, and done/terminal flags.
- Create an environment only for spaces/evaluation, not online data collection.
- Create a
CQN/ILQL-style population withcreate_population(...). - Fill
ReplayBufferwithTransition(...).to_tensordict()entries. - Configure HPO and call
train_offline(...)or a custom offline loop.
Contextual Bandit Flow
- Prepare tabular or array features and targets.
- Create
BanditEnv(features, targets). - Use
NeuralUCBorNeuralTSwith a discrete action space. - Use replay memory and
train_bandits(...)or a custom one-step loop.
Safe Validation
python scripts/inspect_bandit_setup.py --rows 8 --features 4 --arms 3
The helper creates synthetic in-memory data. It does not download UCI datasets or run training.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).