treeple.datasets.approximate_clf_mutual_information_with_monte_carlo#

treeple.datasets.approximate_clf_mutual_information_with_monte_carlo(means, covs, n_samples=100000, class_probs=[0.5, 0.5], base=np.float64(2.718281828459045), seed=None)[source]#

Approximate MI for multivariate Gaussian for a classification setting.

Parameters:

meanslist of ArrayLike of shape (n_dim,): A list of means to sample from for each class.
covslist of ArrayLike of shape (n_dim, n_dim): A list of covariances to sample from for each class.
n_samplesint: The total number of simulation samples
class_probslist, optional: List of class probabilities, by default [0.5, 0.5] for balanced binary classification.
basefloat, optional: The bit base to use, by default np.exp(1) for natural logarithm.
seedint, optional: Random seed for the multivariate normal, by default None.

Returns:

I_XYfloat: Estimated mutual information.
H_Yfloat: Estimated entropy of Y, the class labels.
H_Y_on_Xfloat: The conditional entropy of Y given X.