successor_feature_model
Compute and validate successor-feature reward decompositions for shared-dynamics reinforcement-learning tasks with changing linear rewards.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- successor_feature_model
- description
- Compute and validate successor-feature reward decompositions for shared-dynamics reinforcement-learning tasks with changing linear rewards.
02 资源文件 2
- 脚本scripts/successor_features.py5.5 KB
- 脚本tests/test_successor_features.py2.0 KB
03 SKILL.md 正文
Successor Feature Reward Decomposition
Use this skill when an RL task family shares transition dynamics but changes rewards through a linear feature-weight vector. Do not use it for tasks where rewards cannot be represented or approximated as phi(s,a,s')^T w.
Inputs
- Finite states and actions.
- Deterministic or stochastic transition entries with transition feature vectors.
- A fixed deterministic policy or action-probability table.
- Discount factor
gamma. - Optional reward weight vectors for value reweighting.
Outputs
- Successor-feature table
psi[state][action][feature]. - Dot-product action values for each supplied reward vector.
- Bellman residual and iteration diagnostics.
Workflow
- Normalize transitions into expected feature and next-state probabilities.
- For each policy, solve
psi(s,a) = E[phi(s,a,s') + gamma psi(s', pi(s'))]by bounded value iteration. - Keep reward weights out of the successor-feature update.
- Convert successor features to task values only through
psi^T w. - Reject shape mismatches, missing state-action pairs, and non-convergent updates unless explicitly running a diagnostic.
Validation
Run:
python scripts/successor_features.py --self-test
python tests/test_successor_features.py
Limitations
This skill is intended for small finite recovery experiments or as an exact reference implementation. Large continuous tasks require function approximation not implemented here.
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).