In supervised learning, the generative model and the discriminative model have their own advantages and disadvantages, which are suitable for learning problems under different conditions.

The characteristics of the generative model: the generative model can restore the joint probability distribution P(X, Y), but the discriminative model cannot; the learning convergence speed of the generative model is faster, that is, when the sample size increases, the learned model can be faster The ground converges to the real model; when there are hidden variables, the generative model can still be used for learning, and the discriminative model cannot be used at this time.

The characteristics of the discriminative model: the discriminative model directly learns the conditional probability P(Y|X) or the decision function f(X), and directly faces the prediction, and the accuracy of learning is often higher; because of the direct learning P(Y|X) or f(X) can abstract the data to various degrees, define features and use features, so the learning problem can be simplified.