dqmeasure

Data-quality measures from ISO/IEC 25024 and ISO/IEC 5259-2, implemented as scikit-learn-style estimators.

A measure is scoped to a single column or to the whole table and learns a reference from a clean (train) instance of a table, then measures a dirty (test) instance:

from dqmeasure import DataAccuracyRange, RecordCompleteness

measure = DataAccuracyRange("temperature").fit(train)
units   = measure.predict(test)  # per-cell condition results, a series
x       = measure.score(test)    # the ISO quality measure value, one float

x_table = RecordCompleteness().score(test)  # a table-scoped measure takes no column

Every measure scores higher is better: where a standard defines X in the opposite direction, we report 1 − X. Otherwise X follows the standard, with opinionated defaults where it leaves room for interpretation. The conceptual model is described in the measure model.

The core is written against Narwhals, so inputs may be Polars or pandas frames; results come back in the caller's backend.

See the API reference for the full estimator surface.