Skip to content

API reference

These pages are generated from the source docstrings by mkdocstrings (static analysis — building the docs never imports mli). Public names are re-exported from the top-level mli package, e.g. from parleml_mli import fit_profiler.

In short

  • Two entry points: fit_profiler for a supported model, monitor_predictions for any other scorer or for predictions alone.
  • Both return a ProfilingResult; the other pages document what it holds.
  • Tuning lives in one ProfilingConfig; see Configuration.

Entry points

Page Symbol Purpose
fit_profiler parleml_mli.fit_profiler Profile a natively supported fitted model → ProfilingResult
monitor_predictions parleml_mli.monitor_predictions Model-agnostic bring-your-own-predictions / lite ingest
ProfilingResult parleml_mli.ProfilingResult The central result hub + per-case justifier / justify
Rule extraction parleml_mli.extract_rules Mine IF-THEN rules from a tree ensemble
Global explanations parleml_mli.explain PDP / importance / drift / fairness / …
EDA & leakage parleml_mli.compute_eda Data-level analysis + leakage scan
Evaluation parleml_mli.evaluation.* Holdout metrics, calibration, business value
Trust summary parleml_mli.compute_trust_summary 9-tile trust hero
Model card parleml_mli.autodoc.build_model_card Canonical ModelCard JSON
Local sensitivity parleml_mli.compute_ice_for_case, … Per-case ICE / LOCO / counterfactual
Core types & enums parleml_mli.core.* TaskType, SpaceType, EncodingRegistry, …

Configuration

fit_profiler's tuning knobs are best passed as one typed ProfilingConfig — composed from small section dataclasses (ClusteringConfig, RulesConfig, ExplanationsConfig, …), all re-exported from the top-level mli package. The Configuration (ProfilingConfig) page documents every section and field.

Additional records — EDAConfig (also the ProfilingConfig.eda section), RunSettings and CardinalityCaps — are documented field-by-field on the Configuration parameters page, generated straight from the source.

Next