Regular prediction

t-boost provides the following methods for applying a trained model. For a polars input, the features are matched by name: extra columns are ignored and the order of the columns does not matter.

Classes

TBoostRegressor

Method predict

Description

Apply the model to the given dataset. For the poisson, gamma and tweedie objectives the result is the rate per unit of exposure.

Method predict_raw

Description

Apply the model to the given dataset and return the raw score on the link scale.

TBoostClassifier

Method predict

Description

Apply the model to the given dataset to predict the class labels.

Method predict_proba

Description

Apply the model to the given dataset to predict the probability that the object belongs to the given classes.

Method decision_function

Description

Apply the model to the given dataset and return the raw score on the logit scale.

Usage examples

rate = model.predict(test_data)
expected_claims = rate * test_data["Exposure"].to_numpy()

Use the required_columns attribute to read only the columns the model needs:

import polars as pl

data = pl.scan_parquet("policies.parquet").select(model.required_columns).collect()
preds = model.predict(data)