Which distributions does GaussianNB fit: one Gaussian for the entire feature matrix or separate ones?
Classic MLLevel: IntermediateClass-specific feature distributions
Which distributions does GaussianNB fit: one Gaussian for the entire feature matrix or separate ones?
Hints
- Index parameters by both class and feature.
- Independence removes off-diagonal covariance terms.
Answer
Separate univariate Gaussian feature distributions for each class, using class-specific means and variances. The naive factorization omits arbitrary within-class feature covariance.
Source and licence: scikit-learn 1.7.2 — naive bayes · BSD-3-Clause · answer: Quantoros editorial