Discriminant model and generative model in machine learning-machine learning

What are two models?

We have come to these two concepts from a few words:
1, machine learning is divided into supervised machine learning and unsupervised machine learning;
2, supervised machine learning is known as training set data categories to train the classifier, unsupervised machine learning is not aware of the category of training sets to train the classifier;
3, so that the supervised machine learning can be abstracted as a task of classification, and the basic completion of unsupervised is clustering;
4, supervised machine learning, we can be summarized as through a lot of tagged data, training a model, and then use this, the input of x to predict the output of Y. This model generally has two kinds:

Decision function: Y=f (X)
Conditional probability distribution: P (y| X

5, according to the learning data to obtain the two models of the method, we can be divided into discriminant methods and generation methods;

6, the concept of formal introduction

Discriminant method: Directly from the data learning decision function y=f (X) or conditional probability distribution P (y| X) as a predictive model, i.e. discriminant model. The discriminant approach is concerned with what output y should be predicted for a given input x.

Data Direct Learning decision function y=f (X) or conditional probability distribution P (y| X) is the predictive model, which is the discriminant model.

Generation method: By Data learning Joint probability distribution P (x,y), then by P (y| x) =p (x,y)/p (x) to find the probability distribution P (y| X) as a predictive model. The method represents the generation relationship between the given input x and the output y

P (y| X) as the model of prediction is the generation model; two examples of models

Generation model: Naive Bayesian, Hidden Markov (EM algorithm)
Discriminant models: K-Nearest Neighbor method, Perceptron, decision tree, logistic regression, linear regression, maximum entropy model, support vector Machine (SVM), lifting method, Conditional random field (CRF) comparison

1, the generation model can restore the joint probability distribution (restore the similarity of the data itself), and the discriminant method can not;
2, the learning convergence speed of the generation method is faster, when the sample capacity increases, the model learned can converge to the real model faster;
3, when there are implicit variables, can still use the generation method to learn, at this time the discriminant method can not be used;
4, discriminant learning can not reflect the characteristics of the training data itself, but it looks for the best classification between different categories, reflecting the difference between heterogeneous data, directly facing the prediction, often the higher the accuracy of learning, due to direct learning P (y| x) or y=f (x) so that learning can be simplified;
5. Simply put, the generation model is to find the law from a large number of data, which belongs to statistical learning, and the discriminant model only cares about the difference of different types of data and uses the difference to classify.

：
http://blog.csdn.net/zouxy09/article/details/8195017
Http://www.cnblogs.com/kaituorensheng/p/3379170.html
http://www.zhihu.com/question/20446337