In "Data Science Interview Preparation(#DAY 02).pdf", on page no. 15, you have written like, "each feature are assumed to be distributed according to a Gaussian Distribution but In GNB, we will assume likelihoods to be Gaussian distributed right, not features(not P(x_i), it is P(x_i/y)). and then we will use maximum likelihood estimation to estimate all mean and std of each feature per class, that is why we will get the "sigma_" shape of (n_classes, n_features) in SKlearn implementation as well.
In "Data Science Interview Preparation(#DAY 02).pdf", on page no. 15, you have written like, "each feature are assumed to be distributed according to a Gaussian Distribution but In GNB, we will assume likelihoods to be Gaussian distributed right, not features(not P(x_i), it is P(x_i/y)). and then we will use maximum likelihood estimation to estimate all mean and std of each feature per class, that is why we will get the "sigma_" shape of (n_classes, n_features) in SKlearn implementation as well.