nami_graph_inversion
Generate natural minimally faithful inverse graph structures from Bayesian-network DAGs using the paper's NaMI variable-elimination procedure.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- nami_graph_inversion
- description
- Generate natural minimally faithful inverse graph structures from Bayesian-network DAGs using the paper's NaMI variable-elimination procedure.
02 资源文件 2
- 脚本scripts/nami.py7.2 KB
- 脚本tests/test_nami.py1.8 KB
03 SKILL.md 正文
NaMI Graph Inversion
Use this skill when a task needs the structural inversion method from "Faithful Inversion of Generative Models for Effective Amortized Inference": converting a generative Bayesian-network DAG into inverse parent sets for an amortized inference network. Do not use it for numeric posterior inference by itself; it produces graph structure and an elimination trace, not trained neural parameters.
Inputs
- A JSON object or Python mapping
parentsfrom node id to a list of directed parents in the generative BN. - A list of
latentsthat the inverse network must sample. - A list of
observedvariables that condition the inverse network. mode, eithertopologicalfor forward-NaMI orreverse_topologicalfor reverse-NaMI.
Outputs
inverse_parents: mapping from latent variable to parent variables in the inverse graph.elimination_order: latent variables in the order simulated by variable elimination.trace: per-step frontier, fill edges, unmarked neighbors, and selected variable.edge_count: number of inverse parent edges.
Workflow
- Validate that the generative graph is acyclic and that all latent/observed variables exist.
- Moralize the graph by dropping directions and connecting all co-parents.
- Initialize the frontier with latent variables that have no upstream latent variables for the selected mode.
- Repeatedly choose the frontier variable with the smallest min-fill score, using lexical order for ties.
- Add fill edges among unmarked neighbors, assign those neighbors as inverse parents, mark the variable, and update the frontier.
- Return the inverse structure and trace. Run both modes when compactness matters.
Validation
Run:
python scripts/nami.py --example binary_tree --mode topological
python tests/test_nami.py
The test file uses only the Python standard library and can also be run by the Distiller skill-tree validator.
Limitations
This skill intentionally implements the graph algorithm only. It does not train inference networks, compute analytic Gaussian posteriors, or prove global optimality among all possible minimal I-maps.
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).