recommenders
Use Microsoft Recommenders to prepare recommendation data, train and choose recommender models, evaluate offline metrics, tune experiments, and plan optional Spark/GPU/cloud workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- recommenders
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use Microsoft Recommenders to prepare recommendation data, train and choose recommender models, evaluate offline metrics, tune experiments, and plan optional Spark/GPU/cloud workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/package-overview.md3.1 KB
- 参考文档references/repo-provenance.md1.4 KB
- 参考文档references/repo-routing-metadata.json334 B
- 参考文档references/troubleshooting.md2.8 KB
- 脚本scripts/check_recommenders_environment.py3.8 KB
03 SKILL.md 正文
Microsoft Recommenders
Use this repo skill when a task names recommenders, Microsoft/Recommenders, or needs package-specific guidance for recommendation-system data preparation, model selection/training/scoring, offline metrics, tuning, benchmarks, or operationalization patterns.
Backend truth for this skill
- Verified in the base CPU scope: package import, pandas data utilities/splitters, Python evaluation metrics, SAR, TF-IDF utilities, Cornac/LightGBM helper imports, parameter sweep, K8s sizing utilities, and bundled tiny smoke scripts.
- Optional and not verified in this CPU scope: Spark/PySpark workflows, TensorFlow/PyTorch deep-learning/GPU models, AzureML, Databricks, AKS, SARplus, and experimental dependencies.
- A visible GPU or Spark host is not enough. Verify the matching package extra, framework, system runtime, data, and credentials before claiming backend coverage.
Installation quick start
For ordinary CPU package use:
pip install recommenders
python - <<'PY'
import recommenders
print(recommenders.__version__)
PY
Optional workflow families use documented extras such as recommenders[gpu], recommenders[spark], or recommenders[experimental]. Install only the extra required by the selected workflow; avoid broad [all] installs for small tasks.
Route map
- data-preparation: load or validate interaction data, split train/test data, sample negatives, convert sparse/LibFFM formats, and handle dataset download/cache issues.
- modeling: choose model families, instantiate, fit, score, and recommend with SAR, TF-IDF, Cornac, LightGBM helpers, optional Spark/deep-learning/news/sequential/experimental models, and run tiny model smokes.
- evaluation: compute rating, ranking, diversity, novelty, serendipity, and optional Spark metrics; fix column/type/top-k metric failures.
- operations-and-tuning: plan parameter sweeps, NNI/AzureML tuning, Databricks/AKS operationalization, benchmark comparisons, and environment/backend readiness.
Shared references and scripts
- Read package-overview.md for module families, extras, and workflow-to-sub-skill mapping.
- Read troubleshooting.md for cross-cutting install/import, optional dependency, backend, cloud, and data-download failures.
- Read repo-provenance.md before deciding whether this skill is stale for a different checkout or package version.
- Run check_recommenders_environment.py to check base imports and optionally run all bundled CPU smoke helpers from an installed skill tree.
Suggested workflow order
- Start with
data-preparationto validate columns, split strategy, and candidate generation. - Use
modelingto choose the smallest model family that matches the data and backend. - Use
evaluationto select metrics and validate prediction dataframe contracts. - Use
operations-and-tuningonly when the user needs tuning, benchmarks, deployment, or backend readiness beyond a local CPU smoke.
Safety boundaries
- Do not run dataset downloads, notebooks, cloud scripts, Spark jobs, GPU training, benchmark loops, or cluster mutations unless the user explicitly authorizes the needed network, credentials, compute, and time budget.
- Do not treat skipped optional native cases as passing. Report them as optional or unverified.
- Runtime instructions in this skill are self-contained; do not ask a future agent to open the original source checkout for normal package usage.
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).