knn.zip_人工智能_人工智能导论_人工智能导论作业_水杯分类

import random import numpy as np import matplotlib.pyplot as plt import pickle,pprint from classifier.knn import KNearestNeighbor from classifier.svm import LinearSVM from classifier.softmax import Softmax from classifier.mlp import MultiLayerPerceptron print "hello nut" ##load dataset pkl_file = open('dataset/train_forstu.pickle', 'rb') train = pickle.load(pkl_file) #pprint.pprint(train) pkl_file.close() train_x=np.array(train[0]) train_y=np.array(train[1]) print train_x.shape print train_y.shape ############################################################################ ######### type in the path of test dataset in the next line ########### ############################################################################ pkl_file = open('dataset/valid_forstu.pickle', 'rb') ############################################################################ ######### type in the path of test dataset in the above line ########### ############################################################################ valid = pickle.load(pkl_file) pkl_file.close() valid_x=valid[0] valid_y=valid[1] print valid_x.shape, valid_x.min() print valid_y.shape ,valid_y.min() ############################################################################ ######### type in the model in the next line ########### ############################################################################ model='mlp' #can be 'knn','svm','softmax','mlp' ############################################################################ ######### type in the model in the above line ########### ############################################################################ if model=='knn': classifier = KNearestNeighbor() classifier.train(train_x, train_y) pred_y=classifier.predict(valid_x,k=10) num_correct = np.sum(pred_y == valid_y) accuracy = float(num_correct) / valid_y.shape[0] print 'Got %d / %d correct using knn => accuracy: %f' % (num_correct, valid_y.shape[0], accuracy) result_file = open('predict/knn_predict.txt', 'w') for i in xrange(valid_y.shape[0]): result_file.write(str(pred_y[i])+'\n') result_file.close( ) elif model=='svm': mean_x = np.mean(train_x, axis=0) print "mean_x",mean_x.shape train_x-=mean_x valid_x-=mean_x train_x= np.hstack([train_x, np.ones((train_x.shape[0], 1))]) valid_x= np.hstack([valid_x, np.ones((valid_x.shape[0], 1))]) print train_x.shape,valid_x.shape classifier = LinearSVM() history1=classifier.train(train_x, train_y, learning_rate=1e-7, reg=5e4, num_iters=1000,batch_size=200, verbose=True) history2=classifier.train(train_x, train_y, learning_rate=1e-9, reg=5e4, num_iters=1000,batch_size=200, verbose=True) y_train_pred = classifier.predict(train_x) print 'training accuracy: %f' % (np.mean(train_y == y_train_pred), ) y_val_pred = classifier.predict(valid_x) print 'validation accuracy: %f using svm' % (np.mean(valid_y == y_val_pred), ) result_file = open('predict/svm_predict.txt', 'w') for i in xrange(valid_y.shape[0]): result_file.write(str(y_val_pred[i])+'\n') result_file.close( ) plt.plot(history1) plt.xlabel('Iteration number') plt.ylabel('Loss value') plt.show() elif model=='softmax': mean_x = np.mean(train_x, axis=0) print "mean_x",mean_x.shape train_x-=mean_x valid_x-=mean_x train_x= np.hstack([train_x, np.ones((train_x.shape[0], 1))]) valid_x= np.hstack([valid_x, np.ones((valid_x.shape[0], 1))]) print train_x.shape,valid_x.shape classifier = Softmax() history1=classifier.train(train_x, train_y, learning_rate=1e-7, reg=5e4, num_iters=2000,batch_size=200, verbose=True) #history2=classifier.train(train_x, train_y, learning_rate=1e-9, reg=5e4, num_iters=1000,batch_size=200, verbose=True) y_train_pred = classifier.predict(train_x) print 'training accuracy: %f' % (np.mean(train_y == y_train_pred), ) y_val_pred = classifier.predict(valid_x) print 'validation accuracy: %f using softmax' % (np.mean(valid_y == y_val_pred), ) result_file = open('predict/softmax_predict.txt', 'w') for i in xrange(valid_y.shape[0]): result_file.write(str(y_val_pred[i])+'\n') result_file.close( ) plt.plot(history1) plt.xlabel('Iteration number') plt.ylabel('Loss value') plt.show() elif model=='mlp': #mean_x = np.mean(train_x, axis=0) #train_x-=mean_x #valid_x-=mean_x #train_x/=255 #valid_x/=255 train_y1=np.zeros((train_y.shape[0],6),dtype='float32') valid_y1=np.zeros((valid_y.shape[0],6),dtype='float32') for i in xrange(train_y.shape[0]): train_y1[i,int(train_y[i])]=1 for i in xrange(valid_y.shape[0]): valid_y1[i,int(valid_y[i])]=1 #print valid_y print "x ",train_x.shape,valid_x.shape print "y ",train_y.shape,valid_y.shape classifier=MultiLayerPerceptron() pred_y,loss_history_tr,loss_history_va=classifier.run_mlp(train_x,train_y1,valid_x,valid_y1,num_iters=1000,verbose=True) print pred_y.shape print 'validation accuracy: %f using mlp' % (np.mean(valid_y == pred_y), ) result_file = open('predict/mlp_predict.txt', 'w') for i in xrange(valid_y.shape[0]): result_file.write(str(pred_y[i])+'\n') result_file.close( ) plt.plot(loss_history_tr,"g-",label="train accuracy") plt.plot(loss_history_va,"r-.",label="valid accuracy") plt.xlabel('Iteration number') plt.ylabel('accuracy') plt.legend() plt.show() else: print "please check the model type, they should be from 'knn','svm','softmax','mlp'"