Relu 可视化

黄**

# -*- coding: utf-8 -*-
import numpy as np
from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap
from sklearn import datasets
from sklearn.neural_network import MLPClassifier
from matplotlib.animation import FuncAnimation

def MaxMinNormalization(x):
    x = (x - np.min(x)) / (np.max(x) - np.min(x))
    return x

def datasetC(n):
    np.random.seed(0)
    x,y= datasets.make_circles(n_samples=n, factor=0.05, noise=.2)
    X=255*MaxMinNormalization(x)
    Y=y;Y[Y==0]=-1
    return X,Y

def meshgrid():
    a = np.linspace(0,255,256);b = np.linspace(0,255,256)
    [Xa,Yb] = np.meshgrid(a,b)
    points = [point for point in zip(Xa.flat,Yb.flat)]; np.random.shuffle(points)
    P=np.array(points)
    return P

cm_bright = ListedColormap(['Blue', 'Orange'])

if __name__ == '__main__':
    plt.figure(1, figsize=(4, 4))
    X,Y=datasetC(500)
    plt.scatter(X[:,0], X[:,1], s=10, c=Y, cmap=cm_bright)
    plt.xticks();plt.yticks()
    #plt.savefig('D:\VisualNN\Cdataset.png',format='png',transparent=True,dpi=300);plt.close()

    plt.figure(2, figsize=(4, 4))
    X=X-128
    P=meshgrid();P=P-128;t=0.02;P=np.array(P);P=P[0:int(t*65536)]
    clf =MLPClassifier(hidden_layer_sizes=(3,2),activation='relu',solver='lbfgs',learning_rate_init=0.3,max_iter=2000)
    clf.fit(X, Y);print('score = ',clf.score(X, Y))
    pd=clf.predict(P)
    w=clf.coefs_
    b=clf.intercepts_
    plt.scatter(P[:,0], P[:,1], s=10, c=pd, cmap=cm_bright)
    plt.xticks();plt.yticks()
    #plt.savefig('D:\VisualNN\Cpdrelu(3,2).png',format='png',transparent=True,dpi=300);plt.close()

    ip1=np.dot(  P,w[0])+b[0];  op1=ip1.copy();op1[op1<0]=0;
    ip2=np.dot(op1,w[1])+b[1];  op2=ip2.copy();op2[op2<0]=0;
    ip3=np.dot(op2,w[2])+b[2];  op3=ip3.copy();op3[op3<0]=0;

    fig = plt.figure(3, figsize=(4, 4))
    def update(i):
        ax = Axes3D(fig, elev=15, azim=i)
        ax.scatter(ip1[:,0],ip1[:,1],ip1[:,2], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
        return ax
    i=np.hstack((range(0,90,5),range(90,0,-5)))
    animation = FuncAnimation(fig, update, i,  interval=100)
    #animation.save('D:\VisualNN\Crelu1ip(3,2).gif',  writer='imagemagick');plt.close()

    fig = plt.figure(4, figsize=(4, 4))
    def update(i):
        ax = Axes3D(fig, elev=15, azim=i)
        ax.scatter(op1[:,0],op1[:,1],op1[:,2], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
        return ax
    i=np.hstack((range(0,90,5),range(90,0,-5)))
    animation = FuncAnimation(fig, update, i,  interval=100)
    #animation.save('D:\VisualNN\Crelu1op(3,2).gif',  writer='imagemagick');plt.close()

    fig = plt.figure(5, figsize=(4, 4))
    plt.scatter(ip2[:,0], ip2[:,1], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
    #plt.savefig('D:\VisualNN\Crelu2ip(3,2).png',format='png',bbox_inches='tight',transparent=True,dpi=300);plt.close()

    fig = plt.figure(6, figsize=(4, 4))
    plt.scatter(op2[:,0], op2[:,1], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
    plt.ylim(-1200,390)
    #plt.savefig('D:\VisualNN\Crelu2op(3,2).png',format='png',bbox_inches='tight',transparent=True,dpi=300);plt.close()

    fig = plt.figure(7, figsize=(4, 4))
    plt.scatter(ip3[:,0], ip3[:,0], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
    #plt.savefig('D:\VisualNN\Crelu3ip(3,2).png',format='png',bbox_inches='tight',transparent=True,dpi=300);plt.close()

    fig = plt.figure(8, figsize=(4, 4))
    plt.scatter(op3[:,0], op3[:,0], s=5, c=pd, cmap=ListedColormap(['Blue', 'Orange']))
    #plt.savefig('D:\VisualNN\Crelu3op(3,2).png',format='png',bbox_inches='tight',transparent=True,dpi=300);plt.close()