LassoLarsIC

• Implementa un modelo Lasso con el algoritmo LARS y criterios de información.

• Para la estimación de los parámetros se minimiza la función objetivo:

\min_w \frac{1}{2n} ||Xw -y||_2^2 + \alpha * ||w||_1

from sklearn.linear_model import LassoLarsIC

lassoLarsIC = LassoLarsIC(
    # --------------------------------------------------------------------------
    # The type of criterion to use {'aic', 'bic'}
    criterion='aic',
    # --------------------------------------------------------------------------
    # Whether to fit the intercept for this model.
    fit_intercept=True,
    # --------------------------------------------------------------------------
    # The maximum number of iterations.
    max_iter=500,
    # --------------------------------------------------------------------------
    # When set to True, forces the coefficients to be positive.
    positive=False,
    # --------------------------------------------------------------------------
    # The estimated noise variance of the data. If None, an unbiased estimate is
    # computed by an OLS model. However, it is only possible in the case where
    # n_samples > n_features + fit_intercept.
    noise_variance=None,
)

from sklearn.datasets import load_diabetes

X, y = load_diabetes(return_X_y=True)

lassoLarsIC.fit(X, y)
lassoLarsIC.score(X, y)

0.5134098914486394

lassoLarsIC.alpha_

0.045206256469784906

lassoLarsIC.coef_

array([   0.        , -197.75346667,  522.27003779,  297.15393894,
       -103.94552857,    0.        , -223.92409377,    0.        ,
        514.7480026 ,   54.76900516])

lassoLarsIC.intercept_

152.13348416289602

lassoLarsIC.alphas_

array([2.14804358, 2.01202214, 1.02465091, 0.71509814, 0.29441072,
       0.20086946, 0.15602894, 0.04520626, 0.01239262, 0.01151185,
       0.00493726, 0.0029648 , 0.        ])