TunaWrap

A simple wrapper around Optuna for hyperparameter optimization of scikit-learn compatible models.

Note

This is a work in progress.

Usage

TunaWrap supports two approaches for providing models:

Approach 1: Using model (with sklearn's clone)

from tunawrap import TunaWrap
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import make_scorer, accuracy_score

# Define parameters to optimize
parameters = {
    'n_estimators': {'type': 'int', 'low': 10, 'high': 200},
    'max_depth': {'type': 'int', 'low': 2, 'high': 20},
    'min_samples_split': {'type': 'int', 'low': 2, 'high': 10}
}

# Create a base model (set random_state for reproducibility)
model = RandomForestClassifier(random_state=42)

# Create wrapper and optimize
wrapper = TunaWrap(
    model=model,
    parameters=parameters,
    X=X_train,
    y=y_train,
    scorer=make_scorer(accuracy_score),
    folds=5
)

best_params = wrapper.cook(n_trials=100)

Approach 2: Using model_factory (function that creates models)

from tunawrap import TunaWrap
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier

# Define parameters to optimize
parameters = {
    'classifier__n_estimators': {'type': 'int', 'low': 10, 'high': 200},
    'classifier__max_depth': {'type': 'int', 'low': 2, 'high': 20},
    'classifier__min_samples_split': {'type': 'int', 'low': 2, 'high': 10}
}

# Create wrapper and optimize using a model factory
wrapper = TunaWrap(
    model_factory=lambda: Pipeline([
        ('scaler', StandardScaler()),
        ('classifier', RandomForestClassifier(random_state=42))
    ]),
    parameters=parameters,
    X=X_train,
    y=y_train,
    scorer='accuracy',
    folds=5
)

best_params = wrapper.cook(n_trials=100)

Note: You must provide either model or model_factory, but not both.

Parameter Types

• int: Integer parameters with low and high bounds
• float: Float parameters with low and high bounds
• categorical: Categorical parameters with choices list