stellargraph
Routes StellarGraph graph machine learning tasks across graph construction, samplers, TensorFlow Keras models, link prediction, embeddings, graph classification, time series, calibration, ensembles, interpretability, datasets, and Neo4j connector workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- stellargraph
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Routes StellarGraph graph machine learning tasks across graph construction, samplers, TensorFlow Keras models, link prediction, embeddings, graph classification, time series, calibration, ensembles, interpretability, datasets, and Neo4j connector workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/package-overview.md4.2 KB
- 参考文档references/repo-provenance.md3.0 KB
- 参考文档references/repo-routing-metadata.json492 B
- 参考文档references/troubleshooting.md5.8 KB
- 脚本scripts/check_stellargraph_environment.py5.6 KB
03 SKILL.md 正文
StellarGraph Repo Skill
Use this skill when a task involves the StellarGraph Python package or its TensorFlow/Keras graph machine learning workflows. It is a router: read the smallest sub-skill that matches the user's task, then use the bundled references and scripts from that sub-skill.
First checks
- StellarGraph's package metadata targets Python
>=3.6,<3.9; prefer a Python 3.8 environment when installing or reproducing old examples. - Basic install for package use:
python -m pip install stellargraph. - Use
python -m pip install "stellargraph[demos]"only when notebook-style demo dependencies such asgensim,rdflib,numba, Jupyter, or plotting helpers are needed. - Use
python -m pip install "stellargraph[neo4j]"only for Neo4j connector workflows; it still requires a running Neo4j service. - Run
scripts/check_stellargraph_environment.pywhen you need a safe import, TensorFlow/Keras, graph-construction, and optional backend diagnostic before following a workflow. - Read
references/troubleshooting.mdfor Python/TensorFlow/import, optional dependency, dataset-cache, GPU, and Neo4j failures that affect multiple workflows. - Read
references/repo-provenance.mdbefore deciding whether this skill is stale for a newer checkout or package version.
Route by task
Data, graph objects, and datasets
Read sub-skills/graph-data-loading/SKILL.md
when the task is about:
- constructing
StellarGraphorStellarDiGraphobjects from Pandas, NumPy,IndexedArray, or NetworkX; - choosing node/edge IDs, types, features, weights, or directedness;
- converting/querying graph structure or checking graph readiness for ML;
- using built-in dataset loader classes such as
Cora,MovieLens,WN18, orMETR_LAwithout confusing cache/download behavior with model logic.
Walkers, samplers, and Keras generators
Read sub-skills/sampling-generators/SKILL.md
when the task is about random walks, sampled breadth-first neighborhoods,
UnsupervisedSampler, full-batch, sampled mini-batch, padded-graph,
knowledge-graph, or sliding-window generators.
Use this route before model routes when the failure mentions flow,
Sequence, num_samples, n_size, node/link IDs, batch dimensions, sparse
adjacency tensors, or generator/model incompatibility.
Node classification or node regression with GNNs
Read sub-skills/node-classification-gnns/SKILL.md
for supervised or semi-supervised node workflows using GCN, GAT,
GraphSAGE, DirectedGraphSAGE, HinSAGE, ClusterNodeGenerator + GCN,
RGCN, PPNP, APPNP, or the SGC-style linear GCN pattern.
Use sibling routes for upstream graph construction and generator details; this route owns model selection, Keras tensor wiring, targets, and training/evaluation shape decisions for node tasks.
Link prediction, link regression, and knowledge graphs
Read sub-skills/link-prediction-kg/SKILL.md
for edge-split workflows, GraphSAGELinkGenerator, HinSAGELinkGenerator,
FullBatchLinkGenerator, link inference heads, temporal link prediction, and
knowledge graph completion with KGTripleGenerator, ComplEx, DistMult,
RotatE, RotE, or RotH.
Unsupervised embeddings and representation learning
Read sub-skills/embedding-workflows/SKILL.md
for Node2Vec/DeepWalk, Metapath2Vec, Attri2Vec, unsupervised GraphSAGE,
Deep Graph Infomax, GraphWave, Watch Your Step, or extracting embeddings for
scikit-learn/Keras downstream models.
Graph classification and graph time series
Read sub-skills/graph-time-series-workflows/SKILL.md
for graph-level classification with PaddedGraphGenerator,
GCNSupervisedGraphClassification, DeepGraphCNN, or SortPooling, and for
spatio-temporal forecasting with SlidingFeaturesNodeGenerator, fixed
adjacency, and GCN_LSTM.
Model operations, interpretability, and Neo4j
Read sub-skills/model-ops-interpretability/SKILL.md
for:
custom_keras_layerswhen saving/loading Keras models that contain StellarGraph layers;- calibration (
TemperatureCalibration,IsotonicCalibration, expected calibration error, reliability plots); EnsembleandBaggingEnsemblewrappers;- GCN/GAT saliency and integrated-gradient interpretability;
- optional Neo4j connector objects and service-backed generators.
Package map
Read references/package-overview.md when a
user names an algorithm, dataset, module, notebook title, or capability and you
need to pick the correct route before acting.
Boundaries
- This skill teaches package operation, not repository release maintenance or CI pipeline editing.
- Do not run long notebook demos, dataset downloads, Neo4j services, or GPU checks unless the user explicitly wants that external dependency or runtime.
- Prefer safe bundled scripts and tiny synthetic fixtures before attempting full notebook-scale examples.
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).