深度极限学习机_多层极限学习机_ML-ELM_模式识别_matlab源码

# -*- coding: utf-8 -*- """ Created on Sun Aug 18 19:33:52 2019 深度极限学习机（多层极限学习机，ML-ELM的python实现），有需要的加q讨论 @author: KLY qq392503054 """ import numpy as np from deep_elm import * # In[] # Loading in the data from scipy.io import loadmat # Loading data data = loadmat('drivFace600.mat') x=data['input'].astype(np.float32) yy=data['output'] y=np.zeros([600,4],np.float64) for i in range(600): y[i,yy[i]-1]=1 T=np.copy(y) np.random.seed( 0 ) # divide data set a=np.argsort(np.random.rand(1,600)) trX=x[a[0][:400]] trY=T[a[0][:400]] teX=x[a[0][400:600]] teY=T[a[0][400:600]] ELM_AE_hidden_sizes = [100,50, 30] # create 2 ELM_AE, the DELM would be 6400-500-200-50-4 # Since we are training, set input as training data inpX = trX # Create list to hold our elm_aes elm_ae_list = [] # Size of inputs is the number of inputs in the training set input_size = inpX.shape[1] # For each RBM we want to generate for i, size in enumerate(ELM_AE_hidden_sizes): print('Structure of the',i+1,'ELM_AE: ', ' ', input_size, '->', size) elm_ae_list.append(ELM_AE(input_size, size)) #construct elm-ae class input_size = size a=0 # For each RBM in our list for elm_ae in elm_ae_list: a+=1 # Train a new one elm_ae.ELM_AE_train(inpX) #train each elm_ae one by one # Return the output layer inpX = elm_ae.elm_hidden_outpt(inpX)#the input vetor of next elm_ae are equal to the output vetor of last elm_ae print('Train the',a,'ELM_AE--done') # In[] 堆栈DELM 并训练 输出层 Net = DELM(ELM_AE_hidden_sizes, trX, trY) #创建delm Net.load_from_elm_aes(ELM_AE_hidden_sizes, elm_ae_list)#将训练好的elm_ae的权重赋给delm Net.train()#train delm output=Net.predict(teX) output_label=np.argmax(output,axis=1) true_label=np.argmax(teY,axis=1) # accuracy print('Accuracy of test sets is:',np.sum(output_label==true_label)/len(true_label)*100,'%')