Gamma-Leaf Forest Attention

Installation

For the package installation, first install all the requirements and then install the leaf_att_forest package.

$ pip install -r requirements.txt
$ python setup.py install

Usage

The model interface is scikit-learn like, except it is extended with optimize_weights method which can be executed with the same training data as used for an underlying forest training (see example), or with a new data set (see example).

Code example for model instantiation:

from leaf_att_forest import (
    GLAFParams,
    GammaLeafAttentionForest,
    ForestKind,
    TaskType,
)

model = GammaLeafAttentionForest(
    GLAFParams(
        kind=ForestKind.EXTRA,
        task=TaskType.REGRESSION,
        # Gamma-Leaf Attention Forest Parameters
        leaf_tau=1.0,
        leaf_attention=True,
        n_tau=5,
        fit_tree_weights=True,
        # Base forest parameters
        forest=dict(
            n_estimators=200,
            max_depth=None,
            min_samples_leaf=5,
            random_state=12345,
        ),
    )
)

After the underlying forest should be trained:

model.fit(X_train, y_train)

And then weights are optimized:

model.optimize_weights(X_train, y_train)

In order to estimate weights optimization impact scores for model.predict_original(X_val) and model.predict(X_val) could be compared.