python.md

Python Bindings

The Python bindings build from source. The treeboost module exposes:

Area	Classes
Models	GBDTConfig, GBDTModel, MonotonicConstraint
Tuning	AutoTuner, TunerConfig, ParameterSpace, ParamBounds, TuningMode, OptimizationMetric, TaskType, ModelFormat, GridStrategy, EvalStrategy, TrialResult, SearchHistory
Data	DatasetLoader, BinnedDataset, FeatureInfo, FeatureType, DataPipeline, PipelineConfig, PipelineState, HoldoutSplit, KFoldSplit
Encoding	OrderedTargetEncoder, EncodingMap, CountMinSketch, CategoryFilter, CategoryMapping
Losses	MseLoss, PseudoHuberLoss, BinaryLogLoss, MultiClassLogLoss
Prediction	Prediction, ConformalPredictor

AutoModel and UniversalModel are Rust only.

Build

Install maturin and build the python feature from the repository root:

pip install maturin
maturin develop --release --features python

Example

import numpy as np
from treeboost import GBDTConfig, GBDTModel

X = np.random.randn(10000, 20).astype(np.float32)
y = (X[:, 0] + X[:, 1] * 2 + np.random.randn(10000) * 0.1).astype(np.float32)

config = GBDTConfig()
config.num_rounds = 100
config.max_depth = 6
config.learning_rate = 0.1

model = GBDTModel.train(X, y, config)

GBDTModel.train also accepts feature_names and output_dir. Pass output_dir to write model.rkyv and config.json to that directory.

See API.md for the full GBDTConfig and GBDTModel reference.