graph-property-prediction
Routes OGB graph-property workflows for ogbg datasets, molecular graphs, and code2 conversion.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- graph-property-prediction
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes OGB graph-property workflows for ogbg datasets, molecular graphs, and code2 conversion.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md1.3 KB
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.7 KB
- 脚本scripts/py2graph-smoke.py823 B
03 SKILL.md 正文
Graph Property Prediction
Use this subskill for the graph-property family: ogbg-* datasets, molecular
classification/regression, ogbg-code2, ogbg-ppa, and SMILES-to-graph
conversion.
Use this subskill when
- The task names
GraphPropPredDataset,PygGraphPropPredDataset,DglGraphPropPredDataset, orogbg-*. - The task asks for graph-level evaluation metrics such as ROC-AUC, average precision, RMSE, accuracy, or code2 F1.
- The task mentions
smiles2graph,py2graph, molecular datasets, or code to graph conversion. - The task needs the PyG or DGL graph-property wrappers and the corresponding backend packages are installed.
First decisions
- Read
references/workflows.mdfor the graph property task flow and dataset-specific gotchas. - Read
references/api-reference.mdfor the public class names, input/output shapes, and metric names. - Read
references/troubleshooting.mdwhen a loader, evaluator, or conversion helper fails. - If you need a tiny sanity check for code2-style AST conversion, run
scripts/py2graph-smoke.py.
Main workflow
- Pick the dataset name from the exact
ogbg-*catalog. - Decide whether you need the library-agnostic loader or the PyG/DGL wrapper.
- Load the dataset and inspect
get_idx_split(). - Read the evaluator's expected input format before computing metrics.
- For molecular datasets, convert external molecules with
smiles2graphrather than hand-assembling graph dictionaries. - For
ogbg-code2, treat the evaluator input as token sequences, not numeric labels.
Common routing choices
ogbg-mol*-> graph classification/regression with molecule graphs.ogbg-ppa-> graph classification with accuracy.ogbg-code2-> code-to-graph conversion and F1-style sequence evaluation.smiles2graph-> molecule-to-graph helper backed by rdkit.
Optional backend note
PyG and DGL wrappers are optional. If the current environment does not have the matching backend packages, route the task to the library-agnostic loader or ask for the optional dependency rather than assuming the graph-property skill is broken.
What not to do here
- Do not route node/link/LSC dataset questions into this subskill just because they are graph-shaped.
- Do not tell the user to run the original repository's training scripts as the runtime answer.
- Do not depend on the original checkout for the final skill content.
Related references
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).