Training

t-boost provides two classes for training a model. Both accept polars DataFrame and LazyFrame input directly, with the target and other per-object values named by column.

Classes

TBoostRegressor

Class purpose

Training and applying regression models: squared_error, poisson, gamma and tweedie.

Method

fit

TBoostClassifier

Class purpose

Training and applying classification models: logistic for two classes, softmax for three or more.

Method

fit

Usage examples

from t_boost import TBoostRegressor

model = TBoostRegressor(objective="poisson")
model.fit(train_data, "ClaimCount", exposure="Exposure")

The default parameters already include early stopping, bagging, pruning, banding and graduation, so a model is usually trained without tuning. See Training parameters for the full list and Parameter tuning for tips.

A trained model is deterministic: the same data and the same seed give a bit-identical model, whatever the number of threads.