Bayes分类

• 整体代码，在’mnist’,'cifar10’数据集上train。后续补充细节
• minist精度0.88

// bayes classifer
import numpy as np 
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import normalize
from sklearn.metrics import accuracy_score
from cross_validation import Cross_validation
from download import Download
from sklearn.decomposition import PCA
class Bayes:
    def fit(self, x, y):#获得均值，方差和频率
        self.x = x
        self.y = y
        self.classes = np.unique(y)
        self.parameters = {}
        for i , c in enumerate(self.classes):
            X_index_c=x[np.where(y == c)]#
            m= X_index_c.shape[1]
            X_index_c_mean = np.mean(X_index_c, axis=0, keepdims=True)
            X_index_c_cov = np.cov(X_index_c.T)
            parameters = {'mean':X_index_c_mean,'cov':X_index_c_cov,'prior':X_index_c.shape[0]/ x.shape[0]}
            self.parameters['class' + str(c)] = parameters                                  
    def _pdf(self, x, classes):#用多维高斯分布拟合p(x|y)
        mean = self.parameters['class' + str(classes)]['mean']
        cov  = self.parameters['class' + str(classes)]['cov']
        n,d = x.shape
        p = (-np.log(np.linalg.det(cov))/2 
             -np.diagonal(np.dot(x-mean,np.linalg.inv(cov)).dot((x-mean).T))/2)
        print(p.shape)
        return p
    
    def _predict(self, x):#计算每个种类的p(y)p(x|y)
        output = []
        for y in range(self.classes.shape[0]):
            prior = np.log(self.parameters['class' + str(y)]['prior'])
            posterior = self._pdf(x, y)
            prediction = prior + posterior
            output.append(prediction)
        return output
       
    def predict(self,x):#argmax(p(y)p(x|y))就是最终的结果
        output = self._predict(x)
        output = np.reshape(output, (self.classes.shape[0], x.shape[0]))
        prediction = np.argmax(output, axis=0)
        return prediction
     
if __name__=='__main__':
    '''
    x = datasets.load_iris().data
    y = datasets.load_iris().target
    '''    
    list1= ['mnist','cifar10']
    download = Download(list1[0])
    x,y = download._dowenload()[0],download._dowenload()[1]
    x_train,x_test,y_train,y_test = train_test_split(x,y,test_size=0.2)
    # 利用PCA对原始数据进行降维    
    pca = PCA(n_components=10)
    pca.fit(x_train)
    x_train = pca.transform(x_train)
    x_test = pca.transform(x_test)

    clf = Bayes()
    clf.fit(x_train, y_train)#获得均值方差后验概率
    y_pred = clf.predict(x_test)
    accuracy = accuracy_score(y_test, y_pred)
    print('Accuracy:', round(accuracy,2))

    '''
    n =x.shape[1]
    k=10
    cv=Cross_validation(x,y,k)
    D=cv.segmentation()
    clf = Bayes()
    Accuracy=[]
    for i in range(k):
        test_D=D[i,:,:]
        a=D[i:-1,:,:].flatten()
        b=D[0:i,:,:].flatten()
        train_D=np.concatenate((a,b)).reshape(-1,n+1)
        y_train=train_D[:,-1].astype(int)
        y_test=test_D[:,-1].astype(int)
        clf.fit(train_D[:,0:-1],y_train)
        y_pred = clf.predict(test_D[:,0:-1])
        accuracy = accuracy_score(y_test, y_pred)
        Accuracy.append(accuracy)
    print(round(sum(Accuracy)/len(Accuracy),2))
    '''



   # print('x_tarin', x_train.shape,'D0',D[0].shape)    
    #print(test_D1.shape,train_D1.shape)