Examples¶
End-to-end, copy-pasteable recipes for every ml4 workload. Each page synthesizes a small illustrative dataset, configures the real ml4 API, fits, inspects the result, and saves a deployable artifact. The datasets are deliberately tiny so the pattern stays readable — swap in your own frame and the calls are unchanged.
Install extras
The tabular examples need only the base install. The satellite workloads
pull optional dependencies: pip install "parleml-ml4[unsupervised]" (clustering /
anomaly), "parleml-ml4[forecasting]", "parleml-ml4[ltr]". Association rules need no
extra — mlxtend ships in the base install.
Supervised (tabular)¶
| Example | Entry point | What it shows |
|---|---|---|
| Binary classification (churn) | AutoSupervised |
predict_proba, calibrated + threshold-aware predict, evaluate, save/load |
| Multiclass classification | AutoSupervised |
f1_macro, the class-probability frame, per-class decision thresholds |
| Regression (with intervals) | AutoSupervised |
rmse, CQR predict_interval / predict_intervals, the automatic target transform |
| Domain features | AutoSupervised |
wiring raw email / IP / user-agent / address columns through the stateless parsers |
Non-tabular workloads¶
| Example | Entry point | What it shows |
|---|---|---|
| Clustering (segmentation) | ClusteringDetector |
structured + text fusion, cluster labels, profiles and keywords |
| Anomaly detection (fraud) | AnomalyDetector |
PyOD auto consensus vs a single detector, scores + verdict |
| Forecasting (sales) | Forecaster |
a multi-series panel, predict(h=...), per-series winner + intervals |
| Recommendation (next-best-product) | Recommender |
One-vs-Rest ensemble, top-N per entity |
| Learning-to-Rank | LTRRanker |
grouped decision occasions, NDCG@k, top-ranked candidates |
| Association rules (market basket) | AssociationMiner |
frequent itemsets, lift-ranked rules, FDR pruning |
Where to go next¶
Every example links the matching guide for the full option surface — start with Configuration and Supervised training, then the workload guides under Unsupervised, Forecasting, Recommendation, Learning-to-Rank and Association.