SchemaValidator workflows¶

About¶

Training data establishes the schema that new observations must satisfy. SchemaValidator learns names and dtypes during fit, then checks incoming structure and applies the selected dtype policy during transform.

Why use it?¶

Use it to detect prediction-time schema drift, keep required feature order stable, and make conversions a deliberate policy. Strict rejects dtype changes; semi-strict permits defined widening; permissive attempts strict Polars casts.

Examples¶

Input¶

Training: price: Float64, quantity: Int64.

Code¶

import polars as pl

from polars_learn import SchemaValidator

train = pl.DataFrame({"price": [10.0, 20.0], "quantity": [2, 3]})
incoming = pl.DataFrame({"quantity": ["2", "4"], "price": ["15.0", "25.0"]})
validator = SchemaValidator(mode="permissive", enforce_column_order=False).fit(train)
prepared = validator.transform(incoming)
assert prepared.to_dict(as_series=False) == {"price": [15.0, 25.0], "quantity": [2, 4]}
assert dict(prepared.schema) == validator.schema_

Output¶

The assertions in the example verify the output against the implementation. Python examples are executed during documentation validation.

Practical notes¶

Permissive casts may lose information even when successful; choose the policy for your data contract. Missing required columns still fail. Lazy value checks execute at collection, and nullability is not learned. Consult the class reference and casting policy details for the exact widening and precision rules.