dextra

One line of code → a rich table, a multi-panel plot, and a decision sentence.

dextra is a lightweight toolkit layered on top of pandas and scikit-learn. It turns the repetitive first steps of any data analysis — exploration, cleaning, feature engineering, feature selection and baseline modeling — into single, self-explaining calls. It ships a vocabulary of 63 public functions plus 5 scikit-learn-compatible wrapper classes — 68 distinct public callables in all (see the API consistency table) — organised as the load entry layer, nine analytical modules, and the compat wrappers.

The design contract

• One-line invocation — dx.something(df) is enough to get a useful result.
• Rich tabular output — a densely packed metrics table, not a single number.
• Comprehensive visual — a multi-panel figure beside the numbers.
• Decision sentence — a short Decision: line summarising the result.
• Choices, not decisions — method="compare" ranks candidates without silently picking one (in the feature-engineering, selection and modeling families).
• Leakage-safe fit/apply — learn on train only, replay verbatim on new data via a JSON-serialisable params artifact. This is honoured by the feature-engineering, selection and modeling families and by the two leakage-prone cleaning steps handle_missing and clip_outliers; the remaining cleaning helpers are deterministic, stateless one-shot operations.
• Immutability + audit trail — the input frame is never mutated; every operation is logged to df.attrs["dextra_audit"].

The phases

Phase	Module	What it does
11	_loader	load / peek — raw, messy source → typed DataFrame + full disclosure (the entry layer, "Phase 0" of the pipeline)
1	stats, plots	Rich numeric EDA + better default plots
2	stats_advanced	22 statistical helpers (inference, hypothesis tests, diagnostics)
3	cleaning	10 cleaning helpers across the DAMA-DMBOK stages; leakage-safe fit/apply on handle_missing/clip_outliers
4	features	8 leakage-safe feature-engineering functions
5	selection	5 feature-selection functions (Filter / Embedded / Wrapper)
6	modeling	regress, classify, cluster — instant baselines
7	evaluation	confusion_report, roc_pr, residual_analysis, learning_curves — deep model evaluation
8	timeseries	tsdecomp, tsstat, tsfcast — decomposition, stationarity, baseline forecasts
9	report	edareport — one-call self-contained HTML EDA report
10	dashboard	dash — generated interactive Streamlit app
6.5	compat	scikit-learn-compatible wrappers

See Getting started and the API reference.