heterogeneous-graphs
Build and debug PyTorch Geometric heterogeneous graph data, loaders, HeteroConv models, to_hetero conversions, and hetero link-prediction workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- heterogeneous-graphs
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Build and debug PyTorch Geometric heterogeneous graph data, loaders, HeteroConv models, to_hetero conversions, and hetero link-prediction workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/hetero-api.md7.3 KB
- 参考文档references/hetero-workflows.md7.3 KB
- 参考文档references/troubleshooting.md5.7 KB
- 脚本scripts/hetero_metadata_check.py7.1 KB
03 SKILL.md 正文
Heterogeneous Graphs
Use this sub-skill when the task involves PyG HeteroData, typed metadata, heterogeneous mini-batches, HeteroConv, to_hetero, bipartite message passing, or heterogeneous link prediction.
Quick Routing
- Read
references/hetero-api.mdwhen you need exact object shapes, metadata conventions, loader signatures, or model API choices for hetero graphs. - Read
references/hetero-workflows.mdwhen implementing a complete workflow: constructingHeteroData, converting a homogeneous SAGE-style model withto_hetero, hand-writingHeteroConv, using hetero loaders, or preparing link prediction splits. - Read
references/troubleshooting.mdwhen errors mention missing stores, invalid edge triplets, absent reverse edges,to_heterometadata mismatch, or bipartite size inference. - Run
scripts/hetero_metadata_check.py --helpto inspect a safe synthetic validator for tinyHeteroDatametadata and reverse-edge expectations.
Common Starting Points
- Build a typed graph with
data = HeteroData(), then assign node stores such asdata['paper'].xand edge stores such asdata['author', 'writes', 'paper'].edge_index. - Use
node_types, edge_types = data.metadata()as the canonical metadata tuple forto_hetero,HeteroConvplanning, and hetero loader dictionaries. - For mini-batch node tasks, pass hetero neighbor budgets as
num_neighbors={edge_type: [fanout1, fanout2]}and setinput_nodes=('paper', paper_indices). - For link prediction, keep forward and reverse relation names explicit, and pass matching
edge_typesandrev_edge_typestoRandomLinkSplitwhen reverse edges exist. - For model conversion, prefer
SAGEConv((-1, -1), hidden_channels)or other bipartite-capable layers in the homogeneous model before callingto_hetero(model, data.metadata(), aggr='sum').
Boundaries
- For homogeneous
Data, dataset classes, transforms, and basic graph validation, use the siblingdata-and-datasetsskill. - For generic training-loop design, metrics, pooling, and homogeneous GNN architecture choices, use the sibling
gnn-modelingskill. - For distributed graph stores, remote backends, multi-GPU, or service-based sampling, use the sibling
scalable-distributedskill.
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).