本文共 13373 字,大约阅读时间需要 44 分钟。
当你完成这个,你将完成第四周的最后一个编程作业,也是这门课的最后一个编程作业! 您将使用在上一个作业中实现的功能来构建一个深层网络,并将其应用于卡特彼勒和非卡特彼勒分类。希望您会看到相对于之前的逻辑回归实现,准确性有所提高。 完成这项任务后,您将能够: 建立并应用深层神经网络进行监督学习。 让我们开始吧!
1-准备软件包
让我们首先导入您在本次任务中需要的所有包。
import timeimport numpy as npimport h5pyimport matplotlib.pyplot as pltimport scipyfrom PIL import Imagefrom scipy import ndimagefrom dnn_app_utils_v2 import *%matplotlib inlineplt.rcParams['figure.figsize'] = (5.0, 4.0) # set default size of plotsplt.rcParams['image.interpolation'] = 'nearest'plt.rcParams['image.cmap'] = 'gray'%load_ext autoreload%autoreload 2np.random.seed(1)
2 -数据集
您将使用与“逻辑回归作为神经网络”(作业2)中相同的“猫vs非猫”数据集。你建立的模型对猫和非猫图像的分类有70%的测试准确率。希望你的新模型会表现得更好!
问题陈述:您将获得一个数据集(“数据h5”),其中包含:
让我们更加熟悉数据集。通过运行下面的单元格加载数据。
train_x_orig, train_y, test_x_orig, test_y, classes = load_data()print(train_x_orig.shape)print(train_y.shape)print(test_x_orig.shape)print(test_y.shape)print(classes)
结果:
(209, 64, 64, 3)(1, 209)(50, 64, 64, 3)(1, 50)[b'non-cat' b'cat']
下面的代码将向您显示数据集中的图像。请随意更改索引并多次重新运行单元格以查看其他图像。
# Example of a pictureindex = 50plt.imshow(train_x_orig[index])print ("y = " + str(train_y[0,index]) + ". It's a " + classes[train_y[0,index]].decode("utf-8") + " picture.")
结果:
y = 1. It's a cat picture.
# Explore your dataset m_train = train_x_orig.shape[0]num_px = train_x_orig.shape[1]m_test = test_x_orig.shape[0]print ("Number of training examples: " + str(m_train))print ("Number of testing examples: " + str(m_test))print ("Each image is of size: (" + str(num_px) + ", " + str(num_px) + ", 3)")print ("train_x_orig shape: " + str(train_x_orig.shape))print ("train_y shape: " + str(train_y.shape))print ("test_x_orig shape: " + str(test_x_orig.shape))print ("test_y shape: " + str(test_y.shape))
结果:
Number of training examples: 209Number of testing examples: 50Each image is of size: (64, 64, 3)train_x_orig shape: (209, 64, 64, 3)train_y shape: (1, 209)test_x_orig shape: (50, 64, 64, 3)test_y shape: (1, 50)
像往常一样,在将图像输入网络之前,您需要对图像进行整形和标准化。代码在下面的单元格中给出。
# Reshape the training and test examples train_x_flatten = train_x_orig.reshape(train_x_orig.shape[0], -1).T # The "-1" makes reshape flatten the remaining dimensionstest_x_flatten = test_x_orig.reshape(test_x_orig.shape[0], -1).T# Standardize data to have feature values between 0 and 1.train_x = train_x_flatten/255.test_x = test_x_flatten/255.print ("train_x's shape: " + str(train_x.shape))print ("test_x's shape: " + str(test_x.shape))
结果:
train_x's shape: (12288, 209)test_x's shape: (12288, 50)
12,288等于64×64×3,这是一个重新整形的图像向量的大小。
3 -模型的架构
现在您已经熟悉了数据集,是时候构建一个深层神经网络来区分猫图像和非猫图像了。 您将构建两种不同的模型:
然后,您将比较这些模型的性能,并尝试不同的 L 值 让我们看看这两种架构。
3.1 -两层的神经网络
该模型可以概括为: INPUT -> LINEAR -> RELU -> LINEAR -> SIGMOID -> OUTPUT
3.2 - L层深层神经网络
有一个简化的网络表示:
3.3 -一般方法 像往常一样,您将遵循深度学习方法来构建模型:
a.正向传播
b.计算成本函数
c.反向传播
d.更新参数(使用参数和来自背板的梯度)
4 -两层神经网络
问题:使用您在前面的作业中实现的辅助函数,构建一个具有以下结构的两层神经网络:LINEAR -> RELU -> LINEAR -> SIGMOID。您可能需要的功能及其输入是:
def initialize_parameters(n_x, n_h, n_y): ... return parameters def linear_activation_forward(A_prev, W, b, activation): ... return A, cachedef compute_cost(AL, Y): ... return costdef linear_activation_backward(dA, cache, activation): ... return dA_prev, dW, dbdef update_parameters(parameters, grads, learning_rate): ... return parameters
代码部分:
### CONSTANTS DEFINING THE MODEL ####n_x = 12288 # num_px * num_px * 3n_h = 7n_y = 1layers_dims = (n_x, n_h, n_y)
# GRADED FUNCTION: two_layer_modeldef two_layer_model(X, Y, layers_dims, learning_rate = 0.0075, num_iterations = 3000, print_cost=False): """ 实现一个两层的神经网络,【LINEAR->RELU】 -> 【LINEAR->SIGMOID】 参数: X - 输入的数据,维度为(n_x,例子数) Y - 标签,向量,0为非猫,1为猫,维度为(1,数量) layers_dims - 层数的向量,维度为(n_y,n_h,n_y) learning_rate - 学习率 num_iterations - 迭代的次数 print_cost - 是否打印成本值,每100次打印一次 isPlot - 是否绘制出误差值的图谱 返回: parameters - 一个包含W1,b1,W2,b2的字典变量 """ np.random.seed(1) grads = {} costs = [] # to keep track of the cost m = X.shape[1] # number of examples (n_x, n_h, n_y) = layers_dims # Initialize parameters dictionary, by calling one of the functions you'd previously implemented ### START CODE HERE ### (≈ 1 line of code) parameters = initialize_parameters(n_x, n_h, n_y)# print(parameters.keys())# print(parameters) ### END CODE HERE ### # Get W1, b1, W2 and b2 from the dictionary parameters. W1 = parameters["W1"] b1 = parameters["b1"] W2 = parameters["W2"] b2 = parameters["b2"] # Loop (gradient descent) for i in range(0, num_iterations): # 前向传播: LINEAR -> RELU -> LINEAR -> SIGMOID. Inputs: "X, W1, b1". Output: "A1, cache1, A2, cache2". ### START CODE HERE ### (≈ 2 lines of code) A1, cache1 = linear_activation_forward(X, W1, b1, "relu") A2, cache2 = linear_activation_forward(A1, W2, b2, "sigmoid") ### END CODE HERE ### # 计算代价 ### START CODE HERE ### (≈ 1 line of code) cost = compute_cost(A2,Y) ### END CODE HERE ### # 初始化反向传播 dA2 = - (np.divide(Y, A2) - np.divide(1 - Y, 1 - A2)) # 反向传播. Inputs: "dA2, cache2, cache1". Outputs: "dA1, dW2, db2; also dA0 (not used), dW1, db1". ### START CODE HERE ### (≈ 2 lines of code) dA1, dW2, db2 = linear_activation_backward(dA2, cache2, "sigmoid") dA0, dW1, db1 = linear_activation_backward(dA1, cache1, "relu") ### END CODE HERE ### # Set grads['dWl'] to dW1, grads['db1'] to db1, grads['dW2'] to dW2, grads['db2'] to db2 grads['dW1'] = dW1 grads['db1'] = db1 grads['dW2'] = dW2 grads['db2'] = db2 # 更新参数 ### START CODE HERE ### (approx. 1 line of code) parameters = update_parameters(parameters,grads,learning_rate) ### END CODE HERE ### # Retrieve W1, b1, W2, b2 from parameters W1 = parameters["W1"] b1 = parameters["b1"] W2 = parameters["W2"] b2 = parameters["b2"] # Print the cost every 100 training example if print_cost and i % 100 == 0: print("Cost after iteration {}: {}".format(i, np.squeeze(cost))) if print_cost and i % 100 == 0: costs.append(cost) # 绘制代价 plt.plot(np.squeeze(costs)) plt.ylabel('cost') plt.xlabel('iterations (per tens)') plt.title("Learning rate =" + str(learning_rate)) plt.show() return parameters
测试:
parameters = two_layer_model(train_x, train_y, layers_dims = (n_x, n_h, n_y), num_iterations = 2500, print_cost=True)
结果:
Cost after iteration 0: 0.693049735659989Cost after iteration 100: 0.6464320953428849Cost after iteration 200: 0.6325140647912678Cost after iteration 300: 0.6015024920354665Cost after iteration 400: 0.5601966311605748Cost after iteration 500: 0.515830477276473Cost after iteration 600: 0.4754901313943325Cost after iteration 700: 0.43391631512257495Cost after iteration 800: 0.40079775362038866Cost after iteration 900: 0.3580705011323798Cost after iteration 1000: 0.3394281538366413Cost after iteration 1100: 0.3052753636196264Cost after iteration 1200: 0.27491377282130164Cost after iteration 1300: 0.24681768210614857Cost after iteration 1400: 0.19850735037466094Cost after iteration 1500: 0.17448318112556632Cost after iteration 1600: 0.17080762978096875Cost after iteration 1700: 0.11306524562164708Cost after iteration 1800: 0.09629426845937152Cost after iteration 1900: 0.08342617959726867Cost after iteration 2000: 0.07439078704319087Cost after iteration 2100: 0.06630748132267936Cost after iteration 2200: 0.05919329501038175Cost after iteration 2300: 0.053361403485605606Cost after iteration 2400: 0.04855478562877021
幸好你建立了矢量化的实现!否则训练这个可能要花10倍的时间。 现在,您可以使用经过训练的参数对数据集中的图像进行分类。要查看您对训练集和测试集的预测,请运行下面的单元格。
训练集的预测:
predictions_train = predict(train_x, train_y, parameters)
结果:
Accuracy: 1.0
测试集的预测:
predictions_test = predict(test_x, test_y, parameters)
结果:
Accuracy: 0.72
注意:您可能会注意到,在更少的迭代上运行模型(比如1500次)可以在测试集上提供更好的准确性。这被称为“提前停止”,我们将在下一课中讨论它。提前停止是防止过度拟合的一种方法。 恭喜!看来你的两层神经网络比逻辑回归实现(70%,作业第2周)有更好的性能(72%)。让我们看看你是否可以用一个L层模型做得更好。
5 - L层的神经网络
练习:使用您之前实现的帮助函数来构建一个具有以下结构的L层神经网络: [LINEAR -> RELU]××(L-1) -> LINEAR -> SIGMOID。您可能需要的功能及其输入是:
def initialize_parameters_deep(layer_dims): ... return parameters def L_model_forward(X, parameters): ... return AL, cachesdef compute_cost(AL, Y): ... return costdef L_model_backward(AL, Y, caches): ... return gradsdef update_parameters(parameters, grads, learning_rate): ... return parameters
代码部分:
### CONSTANTS ###layers_dims = [12288, 20, 7, 5, 1] # 5-layer model
# GRADED FUNCTION: L_layer_modeldef L_layer_model(X, Y, layers_dims, learning_rate = 0.0075, num_iterations = 3000, print_cost=False):#lr was 0.009 """ 实现一个L层神经网络:[LINEAR-> RELU] *(L-1) - > LINEAR-> SIGMOID。 参数: X - 输入的数据,维度为(n_x,例子数) Y - 标签,向量,0为非猫,1为猫,维度为(1,数量) layers_dims - 层数的向量,维度为(n_y,n_h,···,n_h,n_y) learning_rate - 学习率 num_iterations - 迭代的次数 print_cost - 是否打印成本值,每100次打印一次 isPlot - 是否绘制出误差值的图谱 返回: parameters - 模型学习的参数。 然后他们可以用来预测。 """ np.random.seed(1) costs = [] # keep track of cost # 初始化参数 ### START CODE HERE ### parameters = initialize_parameters_deep(layers_dims) ### END CODE HERE ### # 循环 (梯度下降) for i in range(0, num_iterations): # 前向传播: [LINEAR -> RELU]*(L-1) -> LINEAR -> SIGMOID. ### START CODE HERE ### (≈ 1 line of code) AL, caches = L_model_forward(X,parameters) ### END CODE HERE ### # 计算代价 ### START CODE HERE ### (≈ 1 line of code) cost = compute_cost(AL, Y) ### END CODE HERE ### # 后向传播 ### START CODE HERE ### (≈ 1 line of code) grads = L_model_backward(AL,Y,caches) ### END CODE HERE ### # 更新参数 ### START CODE HERE ### (≈ 1 line of code) parameters = update_parameters(parameters,grads,learning_rate) ### END CODE HERE ### # Print the cost every 100 training example if print_cost and i % 100 == 0: print ("Cost after iteration %i: %f" %(i, cost)) if print_cost and i % 100 == 0: costs.append(cost) # 绘制代价 plt.plot(np.squeeze(costs)) plt.ylabel('cost') plt.xlabel('iterations (per tens)') plt.title("Learning rate =" + str(learning_rate)) plt.show() return parameters
现在,您将把模型训练成一个5层神经网络。 运行下面的单元来训练你的模型。每次迭代的成本都应该降低。运行2500次迭代可能需要5分钟。检查“迭代0后的成本”是否与下面的预期输出匹配,如果不匹配,请单击笔记本上栏的正方形(⬛)来停止单元格并尝试查找错误。
测试:
parameters = L_layer_model(train_x, train_y, layers_dims, num_iterations = 2500, print_cost = True)
结果:
Cost after iteration 0: 0.771749Cost after iteration 100: 0.672053Cost after iteration 200: 0.648263Cost after iteration 300: 0.611507Cost after iteration 400: 0.567047Cost after iteration 500: 0.540138Cost after iteration 600: 0.527930Cost after iteration 700: 0.465477Cost after iteration 800: 0.369126Cost after iteration 900: 0.391747Cost after iteration 1000: 0.315187Cost after iteration 1100: 0.272700Cost after iteration 1200: 0.237419Cost after iteration 1300: 0.199601Cost after iteration 1400: 0.189263Cost after iteration 1500: 0.161189Cost after iteration 1600: 0.148214Cost after iteration 1700: 0.137775Cost after iteration 1800: 0.129740Cost after iteration 1900: 0.121225Cost after iteration 2000: 0.113821Cost after iteration 2100: 0.107839Cost after iteration 2200: 0.102855Cost after iteration 2300: 0.100897Cost after iteration 2400: 0.092878
训练集的预测:
predictions_train = predict(train_x, train_y, parameters)
结果:
Accuracy: 0.9856459330143541
测试集的预测:
predictions_test = predict(test_x, test_y, parameters)
结果:
Accuracy: 0.8
恭喜!在同一测试集中,您的5层神经网络的性能(80%)似乎优于您的2层神经网络(72%)。 这是这项任务的良好表现。干得好! 尽管在下一堂关于“改进深层神经网络”的课程中,您将学习如何通过系统地搜索更好的超参数(学习速率、层亮度、迭代次数以及其他您在下一堂课中也将学习的参数)来获得更高的精度。
6- 结果分析
首先,让我们来看看一些图像,在L层模型中被错误地标记了,导致准确率没有提高。
print_mislabeled_images(classes, test_x, test_y, pred_test)
分析一下我们就可以得知原因了:
模型往往表现欠佳的几种类型的图像包括:
7-用自己的形象测试(可选/未分级练习)
祝贺你完成这项任务。您可以使用自己的图像并查看模型的输出。为此:
%matplotlib inline## START CODE HERE ##my_image = "my_image.jpg" # change this to the name of your image file my_label_y = [1] # the true class of your image (1 -> cat, 0 -> non-cat)## END CODE HERE ##fname = "images/" + my_imageimage = np.array(ndimage.imread(fname, flatten=False))my_image = scipy.misc.imresize(image, size=(num_px,num_px)).reshape((num_px*num_px*3,1))my_predicted_image = predict(my_image, my_label_y, parameters)plt.imshow(image)print ("y = " + str(np.squeeze(my_predicted_image)) + ", your L-layer model predicts a \"" + classes[int(np.squeeze(my_predicted_image)),].decode("utf-8") + "\" picture.")
结果:
Accuracy: 1.0y = 1.0, your L-layer model predicts a "cat" picture.
转载地址:http://lpuvi.baihongyu.com/