mljar-supervised
Use MLJAR AutoML for tabular classification, regression, fairness-aware training, reports, persistence, and generated Mercury apps.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- mljar-supervised
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use MLJAR AutoML for tabular classification, regression, fairness-aware training, reports, persistence, and generated Mercury apps.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/package-overview.md4.1 KB
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json460 B
- 参考文档references/troubleshooting.md4.1 KB
- 脚本scripts/check_mljar_supervised_install.py4.0 KB
03 SKILL.md 正文
MLJAR Supervised
Use this repo skill when a task involves mljar-supervised, supervised.AutoML, MLJAR AutoML, automated tabular machine learning, MLJAR reports, fairness-aware AutoML, or generated Mercury prediction apps.
MLJAR Supervised trains supervised AutoML pipelines for tabular binary classification, multiclass classification, and regression. It handles common preprocessing, model search, reports, explainability outputs, fairness options, persistence, and app generation through the Python package imported as supervised.
Start here
-
Confirm the package is installed and importable:
from supervised import AutoML print(AutoML)For a more detailed local check, run
scripts/check_mljar_supervised_install.py. -
Identify whether the user wants to train a model, prepare data, inspect an existing run, add fairness constraints, or build an app.
-
Route to the focused sub-skill below before giving detailed API, data, report, or deployment instructions.
-
Keep user examples bounded: disable expensive feature engineering and explanations for smoke checks; use larger
Perform,Compete, orOptunaruns only when the user accepts the runtime.
Route map
| User intent | Read |
|---|---|
Train/configure AutoML, choose mode, algorithms, validation, metrics, time limits, ensembling/stacking, predict/score/retrain |
training-core |
Prepare X, y, sample_weight, custom CV, missing/categorical/text/datetime columns, target inference, preprocessing and feature-engineering flags |
data-preprocessing |
Inspect results_path, load saved runs, use leaderboards, reports, structured reports, explainability artifacts, SHAP/importance/learning curves, persistence |
artifacts-reports |
Configure fairness-aware training with sensitive_features, fairness metrics, thresholds, privileged/underprivileged groups, fairness reports |
fairness-workflows |
Generate app files, preview locally with Mercury, or prepare guarded publishing with app(), local_app(), publish_app() |
app-deployment |
Package assumptions
- Distribution name:
mljar-supervised; import name:supervised. - Main public class:
supervised.AutoML. - Supported ML tasks: binary classification, multiclass classification, and regression.
- Built-in modes:
Explain,Perform,Compete, andOptuna. - Common algorithms include
Baseline,Linear,Decision Tree,Random Forest,Extra Trees,LightGBM,Xgboost,CatBoost,Neural Network, andNearest Neighbors. - Core workflows are CPU-compatible. GPU hardware is not required for this skill's selected scope.
- Optional surfaces can need extra system/runtime setup: Graphviz for some tree visualizations and Mercury plus browser/network/auth for local or hosted apps.
Read references/package-overview.md for install prerequisites, high-level capabilities, optional dependencies, and safe smoke-test strategy. Read references/troubleshooting.md for cross-cutting install/import/dependency failures.
Common operating patterns
- For a quick user-data smoke, use
AutoML(mode="Explain", algorithms=["Baseline", "Decision Tree"], explain_level=0, train_ensemble=False, stack_models=False, golden_features=False, features_selection=False, total_time_limit=30)and a disposableresults_path. - For production-style tabular modeling, start from
mode="Perform", set a realistictotal_time_limit, and keepresults_pathstable for later reports and reload. - For competition or maximum-performance work, use
mode="Compete"ormode="Optuna"only after discussing runtime, validation, stacking, and per-algorithm tuning budgets. - For saved runs, load the full AutoML results directory with
AutoML(results_path="..."); do not load learner files directly unless a sub-skill explicitly says why. - For fairness, pass sensitive features to
fit(..., sensitive_features=...)and choose task-compatible fairness metrics. - For generated apps, call
automl.app()to create files first; do not start Mercury or publish without explicit user approval.
Bundled checks
scripts/check_mljar_supervised_install.py: import/version/signature and optional dependency probe.sub-skills/training-core/scripts/mljar_automl_smoke.py: tiny synthetic training/prediction smoke.sub-skills/data-preprocessing/scripts/inspect_preprocessing_behaviour.py: lightweight preprocessing utility checks.sub-skills/artifacts-reports/scripts/structured_report_smoke.py: tiny structured-report and reload smoke.sub-skills/fairness-workflows/scripts/fairness_smoke.py: tiny fairness-aware fit and report signal check.sub-skills/app-deployment/scripts/generate_app_workspace_smoke.py: tiny app-workspace generation check without serving or publishing.
Refresh and provenance
Read references/repo-provenance.md before deciding whether this skill is current for a checkout. If the package version, commit, public AutoML signatures, docs, examples, or test-backed behavior changed, run refresh-repo-skill instead of relying on this snapshot.
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).