Exceptions and warnings¶

All errors raised by t-boost derive from TBoostError:

from t_boost import TBoostError, SerializationError

Exceptions¶

TBoostError¶

The base class of all t-boost errors. Catching it catches every error below.

Invalid input and configuration¶

Malformed input data (for example, a non-finite feature value or a target outside the domain of the objective), arrays whose shapes disagree and parameter values outside their range raise an exception that is both a TBoostError and a built-in ValueError. A buffer of the wrong element type raises one that is both a TBoostError and a TypeError. So except ValueError and except TBoostError both catch them.

The estimators also raise a plain ValueError when a parameter cannot be honoured for the task, for example a regression-only parameter on a multiclassification model.

SerializationError¶

Saving or loading a model failed: for example, a document written by a newer version of t-boost, or metadata that is not JSON-serializable.

ExactnessError¶

An operation that would break the exact decomposition of the model into rating tables was attempted.

InvariantError¶

One of the internal checks of the exact decomposition failed. This indicates a bug: please report it.

InternalError¶

An internal error. This indicates a bug: please report it.

Warnings¶

PrecisionWarning¶

t-boost computes in float32. When a numeric feature is converted to float32 before training or scoring, a PrecisionWarning (a subclass of UserWarning) is issued once per estimator. Cast the columns to float32 to skip the conversion, or filter the warning:

import warnings
from t_boost import PrecisionWarning

warnings.filterwarnings("ignore", category=PrecisionWarning)