Skip to main content

scikit-learn Example

Timber supports scikit-learn tree estimators and pipelines, including StandardScaler fusion.

Supported Estimators​

  • GradientBoostingClassifier / GradientBoostingRegressor
  • RandomForestClassifier / RandomForestRegressor
  • HistGradientBoostingClassifier / HistGradientBoostingRegressor
  • DecisionTreeClassifier / DecisionTreeRegressor
  • Pipeline with StandardScaler + any of the above

Pipeline with StandardScaler​

This is the most powerful use case — Timber fuses the scaler into tree thresholds at compile time, eliminating preprocessing entirely.

import pickle
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

data = load_breast_cancer()
X_train, X_test, y_train, y_test = train_test_split(
    data.data, data.target, test_size=0.2, random_state=42
)

pipe = Pipeline([
    ("scaler", StandardScaler()),
    ("model", GradientBoostingClassifier(
        n_estimators=50, max_depth=3, random_state=42
    )),
])
pipe.fit(X_train, y_train)

with open("sklearn_pipeline.pkl", "wb") as f:
    pickle.dump(pipe, f)

Load and Serve​

timber load sklearn_pipeline.pkl --name sklearn-bc
timber serve sklearn-bc

How Pipeline Fusion Works​

When Timber encounters a Pipeline(StandardScaler, TreeEstimator):

  1. The parser emits a ScalerStage (with means μᵢ and scales σᵢ) followed by a TreeEnsembleStage
  2. The Pipeline Fusion optimizer pass absorbs the scaler into tree thresholds:
    • New threshold: θ' = θ × σ + μ
  3. The scaler stage is removed from the IR
  4. The compiled code performs inference with zero preprocessing overhead

Random Forest​

from sklearn.ensemble import RandomForestClassifier
import pickle

model = RandomForestClassifier(n_estimators=100, max_depth=5, random_state=42)
model.fit(X_train, y_train)

with open("rf_model.pkl", "wb") as f:
    pickle.dump(model, f)
timber load rf_model.pkl --name random-forest