link-prediction-kg
Guides StellarGraph link prediction, link regression, edge splitting, temporal link workflows, and knowledge graph completion with ComplEx, DistMult, RotatE, RotE, and RotH.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- link-prediction-kg
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides StellarGraph link prediction, link regression, edge splitting, temporal link workflows, and knowledge graph completion with ComplEx, DistMult, RotatE, RotE, and RotH.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md3.7 KB
- 参考文档references/knowledge-graphs.md3.0 KB
- 参考文档references/link-workflows.md3.6 KB
- 参考文档references/troubleshooting.md2.5 KB
- 脚本scripts/link_prediction_smoke.py2.4 KB
03 SKILL.md 正文
Link Prediction and Knowledge Graphs
Use this sub-skill when a StellarGraph task predicts edges/ratings/relations, performs link regression/classification, splits graph edges into train/test sets, or scores triples in a knowledge graph.
Read first
references/link-workflows.mdfor edge splitting, link generators, Keras link heads, temporal link prediction, and leakage checks.references/knowledge-graphs.mdforKGTripleGenerator,ComplEx,DistMult,RotatE,RotE,RotH, negative sampling, and ranking.references/api-reference.mdfor verified signatures and link/KG object responsibilities.references/troubleshooting.mdfor edge leakage, invalid endpoints, target-shape, negative sampling, and relation-type failures.scripts/link_prediction_smoke.pyfor a tiny GCN link-classification wiring check.
Route here when the user asks to
- use
EdgeSplitter.train_test_split; - build link prediction with
GraphSAGELinkGenerator,HinSAGELinkGenerator,FullBatchLinkGenerator,Attri2VecLinkGenerator, orNode2VecLinkGenerator; - add
link_classification,link_regression, orlink_inferenceto source and destination node embeddings; - perform knowledge graph completion with
KGTripleGenerator,ComplEx,DistMult,RotatE,RotE, orRotH; - use
SelfAdversarialNegativeSamplingor rank triples against all nodes; - adapt CTDNE or temporal link-prediction workflows.
Route elsewhere
- Raw graph and dataset construction:
../graph-data-loading/SKILL.md. - Generator shape debugging:
../sampling-generators/SKILL.md. - Node classification:
../node-classification-gnns/SKILL.md. - Node embeddings before downstream edge tasks:
../embedding-workflows/SKILL.md.
Operating workflow
- Construct a graph with valid node IDs, edge IDs/types, and features required by the chosen embedding model.
- Split edges for evaluation with
EdgeSplitteror an equivalent leakage-safe method. Keep train and test positives/negatives separate. - Choose a link generator:
- GCN/GAT/PPNP-style full-batch embeddings:
FullBatchLinkGenerator; - GraphSAGE:
GraphSAGELinkGenerator; - HinSAGE:
HinSAGELinkGeneratorwith head node types; - Attri2Vec/Node2Vec link heads: matching link generators;
- knowledge graph completion:
KGTripleGenerator.
- GCN/GAT/PPNP-style full-batch embeddings:
- Build the base embedding model and call
in_out_tensors(). - Attach a link head such as
link_classification,link_regression, orlink_inferenceto combine source/destination embeddings. - Compile and train/evaluate with generator
flow(link_ids, targets, ...). - For knowledge graph models, use negative sampling and ranking routines designed for triples rather than ordinary binary edge pairs.
Safe check
python sub-skills/link-prediction-kg/scripts/link_prediction_smoke.py --help
python sub-skills/link-prediction-kg/scripts/link_prediction_smoke.py
The smoke uses a tiny graph and does not download datasets or train a notebook.
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).