Andrew Ng 深度学习课程Deeplearning.ai 编程作业——deep Neural network for image classification（1-4.2）

1.Package

import numpy as np  #scientific compute packageimport matplotlib.pyplot as plt  #graphs packageimport h5py  #contact with h5 fileimport scipy from scipy import ndimage  #import our image and reshape to the specific sizefrom A_deeper_neural_network import initialize_parameters_deep,linear_activation_forward,compute_cost,linear_activation_backwardfrom A_deeper_neural_network import update_parameters #这里我导入了自己写的文件的模块A_deeper_neural_networkplt.rcParams["figure.figsize"]=(5.0,4.0)  #set the figure figzie to (5.0,4.0)plt.rcParams["image.interpolation"]="nearest"plt.rcParams["image.cmap"]='gray'np.random.seed(1)  #set the random initial seeddef load_dataset():  #定义导入文件的函数    train_dataset=h5py.File ('/home/hansry/python/DL/1-4/assignment4/datasets/train_catvnoncat.h5',"r")    train_set_x_orig = np.array(train_dataset["train_set_x"][:])     train_set_y_orig = np.array(train_dataset["train_set_y"][:])    test_dataset=h5py.File('/home/hansry/python/DL/1-4/assignment4/datasets/test_catvnoncat.h5',"r")    test_set_x_orig = np.array(test_dataset["test_set_x"][:])    test_set_y_orig = np.array(test_dataset["test_set_y"][:])    classes = np.array(test_dataset["list_classes"][:]) # the list of classes    train_set_y_orig=train_set_y_orig.reshape((1,train_set_y_orig.shape[0]))    test_set_y_orig=test_set_y_orig.reshape(1,test_set_y_orig.shape[0])    return train_set_x_orig,train_set_y_orig,test_set_x_orig,test_set_y_orig,classestrain_set_x_orig,train_set_y,test_set_x_orig,test_set_y,classes=load_dataset()index=7plt.imshow(train_set_x_orig[index]) #to show the package plt.show()print("y="+str(train_set_y[0][index])+" It's a "+classes[train_set_y[0][index]].decode("utf-8")+" picture")

Expected output:

2.datasets

train_set_x_orig,train_set_y,test_set_x_orig,test_set_y,classes=load_dataset()num_px=train_set_x_orig.shape[1]train_x_flatten=train_set_x_orig.reshape(train_set_x_orig.shape[0],num_px*num_px*3).Ttest_x_flatten=test_set_x_orig.reshape(test_set_x_orig.shape[0],num_px*num_px*3).Ttrain_x=train_x_flatten/255  #centralize and normorlize the datasetstest_x=test_x_flatten/255print ("train_x's shape:"+str(train_x.shape))print ("test_x's shape:"+str(test_x.shape))

Expected output：

train_x's shape:(12288, 209)test_x's shape:(12288, 50)

3.Architecture of your model

4.Two-layer neural network

n_x=12288n_h=7n_y=1layer_dims=(n_x,n_h,n_y)def two_layer_model(X,Y,layers_dims,learning_rate,num_iterations,print_cost=True):    np.random.seed(1)    parameters=initialize_parameters_deep(layer_dims)    costs=[]    grads={}    W1=parameters["W1"]    b1=parameters["b1"]    W2=parameters["W2"]    b2=parameters["b2"]    for i in range(num_iterations):        A1,cache1=linear_activation_forward(X,W1,b1,activation="relu") #       print cache1[0][0].shape #       print A1        A2,cache2=linear_activation_forward(A1,W2,b2,activation="sigmoid")        cost=compute_cost(A2,Y)#        costs.append(cost)        dA2=-(np.divide(Y,A2)-np.divide(1-Y,1-A2))        dA1,dW2,db2=linear_activation_backward(dA2,cache2,activation="sigmoid")        dA0,dW1,db1=linear_activation_backward(dA1,cache1,activation="relu")        grads["dW2"]=dW2        grads["db2"]=db2        grads["dW1"]=dW1        grads["db1"]=db1        parameters=update_parameters(parameters,grads,learning_rate)        W1=parameters["W1"]        b1=parameters["b1"]        W2=parameters["W2"]        b2=parameters["b2"]        if print_cost and i%100==0:            print ("cost after iteration{}:{}".format(i,np.squeeze(cost)))        if i%100==0:            costs.append(cost)    costs=np.squeeze(costs)    plt.plot(costs)    plt.ylabel("cost")    plt.xlabel("iterations per one thousand")    plt.title("learning rate :"+str(learning_rate))    plt.show()    return parametersparameters=two_layer_model(train_x,train_set_y,layer_dims,learning_rate=0.0075,num_iterations=3000,print_cost=True)

Expected output:

cost after iteration0:0.69304973566cost after iteration100:0.646432095343cost after iteration200:0.632514064791cost after iteration300:0.601502492035cost after iteration400:0.560196631161cost after iteration500:0.515830477276cost after iteration600:0.475490131394cost after iteration700:0.433916315123cost after iteration800:0.40079775362cost after iteration900:0.358070501132cost after iteration1000:0.339428153837cost after iteration1100:0.30527536362cost after iteration1200:0.274913772821cost after iteration1300:0.246817682106cost after iteration1400:0.198507350375cost after iteration1500:0.174483181126cost after iteration1600:0.170807629781cost after iteration1700:0.113065245622cost after iteration1800:0.0962942684594cost after iteration1900:0.0834261795973cost after iteration2000:0.0743907870432cost after iteration2100:0.0663074813227cost after iteration2200:0.0591932950104cost after iteration2300:0.0533614034856cost after iteration2400:0.0485547856288cost after iteration2500:0.0441405969255cost after iteration2600:0.0403456450042cost after iteration2700:0.0368412198948cost after iteration2800:0.0336603989271cost after iteration2900:0.0307555969578

def prediction(parameters,X,Y):  #对训练集和预测集进行精度评判    W1=parameters["W1"]    b1=parameters["b1"]    W2=parameters["W2"]    b2=parameters["b2"]    A1,cache1=linear_activation_forward(X,W1,b1,activation="relu")    A2,cache2=linear_activation_forward(A1,W2,b2,activation="sigmoid")    predictions=(A2>0.5)  #将大于0的设置为1    accuracy_per=float(np.dot(Y,predictions.T)+np.dot(1-Y,1-predictions.T))/float(Y.size)*100    return predictions,accuracy_perpredictions,accuracy=prediction(parameters,train_x,train_set_y)print ("train_accuracy: "+str(accuracy))predictions,accuracy=prediction(parameters,test_x,test_set_y)print ("test_accuracy: "+str(accuracy))

Expected output：

train_accuracy: 100.0test_accuracy: 72.0

Congratulations! It seems that your 2-layer neural network has better performance (72%) than the logistic regression implementation (70%, assignment week 2). Let’s see if you can do even better with an LL-layer model.

5.L-layer Neural Network

layer_dims=[12288,20,7,5,1]def  L_layer_model(X,Y,layer_dims,num_iterations,learning_rate,print_cost):    np.random.seed(1)    parameters=initialize_parameters_deep(layer_dims)    costs=[]    for i in range(num_iterations):              AL,caches=L_model_layer(X,parameters)        cost_L_model=compute_cost(AL,Y)          grads=L_model_backward(AL,Y,caches)          parameter=update_parameters(parameters,grads,learning_rate)        if print_cost and i%100==0:            print ("cost after iterations{}:{}".format(i,cost_L_model))        if i%100==0:            costs.append(cost_L_model)    return parameter,costsdef prediction_L_model(parameters,X,Y):    AL,caches=L_model_layer(X,parameters)        predictions=(AL>0.5)    accuracy=float(np.dot(Y,predictions.T)+np.dot(1-Y,(1-predictions).T))/float(Y.size)*100    return accuracy,predictionsparameters,costs=L_layer_model(train_x,train_set_y,layer_dims,learning_rate=0.01,num_iterations=3000,print_cost=True)costs=np.squeeze(costs)plt.plot(costs)plt.ylabel("cost")plt.xlabel("iterations")plt.show()train_accuracy,train_predictions=prediction_L_model(parameters,train_x,train_set_y)print ("train_accuracy:"+str(train_accuracy)+"%")test_accuracy,test_predictions=prediction_L_model(parameters,test_x,test_set_y)print ("test_accuracy:"+str(test_accuracy)+"%")

Expected output:

train_x's shape:(12288, 209)test_x's shape:(12288, 50)cost after iterations0:0.771749328424cost after iterations100:0.669269663073cost after iterations200:0.638873866746cost after iterations300:0.597884241863cost after iterations400:0.568827182668cost after iterations500:0.461260004201cost after iterations600:0.508483601988cost after iterations700:0.32759554358cost after iterations800:0.31039799625cost after iterations900:0.24883052978cost after iterations1000:0.207309305492cost after iterations1100:0.140485374517cost after iterations1200:0.115670324218cost after iterations1300:0.0992596314732cost after iterations1400:0.0858446278017cost after iterations1500:0.0749750709344cost after iterations1600:0.0678088205921cost after iterations1700:0.0584015277879cost after iterations1800:0.0520540925361cost after iterations1900:0.0476796512902cost after iterations2000:0.0422589466752cost after iterations2100:0.0377972361751cost after iterations2200:0.0347303021461cost after iterations2300:0.0313911132159cost after iterations2400:0.0287875716657cost after iterations2500:0.0264843633988cost after iterations2600:0.0243811278867cost after iterations2700:0.0226565055434cost after iterations2800:0.021282864075cost after iterations2900:0.0196948174932

在这里需要注意的是初始化权重的函数，即 initialize_parameters_deep(layer_dims)，具体内容如下：

def initialize_parameters_deep(layer_dims):    """    Arguments:    layer_dims -- python array (list) containing the dimensions of each layer in our network    Returns:    parameters -- python dictionary containing your parameters "W1", "b1", ..., "WL", "bL":                    Wl -- weight matrix of shape (layer_dims[l], layer_dims[l-1])                    bl -- bias vector of shape (layer_dims[l], 1)    """    np.random.seed(1)    parameters = {}    L = len(layer_dims)            # number of layers in the network    for l in range(1, L):        parameters['W' + str(l)] = np.random.randn(layer_dims[l], layer_dims[l-1]) / np.sqrt(layer_dims[l-1]) #*0.01，注意这里不是乘以0.01，因为会陷入局部极小值        parameters['b' + str(l)] = np.zeros((layer_dims[l], 1))        assert(parameters['W' + str(l)].shape == (layer_dims[l], layer_dims[l-1]))        assert(parameters['b' + str(l)].shape == (layer_dims[l], 1))    return parameters

6.Result Analysis

First, let’s take a look at some images the L-layer model labeled incorrectly. This will show a few mislabeled images.

def mislabeled_images(classes,X,Y,p):    precision=Y+p    print precision    mislabel_indices=np.asarray(np.where(precision==1))    print mislabel_indices    num_image=mislabel_indices.shape[1]    plt.rcParams["figure.figsize"]=(10.0,10.0)    for i in range(num_image):        index=mislabel_indices[1,i]        plt.subplot(num_image//3,3,i+1)        plt.imshow(X[:,index].reshape(64,64,3),interpolation='nearest')        plt.axis=('off')        plt.title("Prediction:"+str(classes[int(p[0,index])].decode("utf-8"))+"\n"+"Real:"+str(classes[Y[0,index]].decode("utf-8")))mislabeled_images(classes,test_x,test_set_y,test_predictions)

如上图所示，在50张图中，9张预测错误，准确率为82%

A few type of images the model tends to do poorly on include:
Cat body in an unusual position
Cat appears against a background of a similar color
Unusual cat color and species
Camera Angle
Brightness of the picture
Scale variation (cat is very large or small in image)