[GitHub]Python Deep Learning

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

本帖隐藏的内容

Python-Deep-Learning-master.zip (4.81 MB, 需要: 1 个论坛币)

2017-7-8 00:05:09 上传

需要: 1 个论坛币  [购买]

1. Python Deep Learning
   
2. Valentino Zocca et al.
   
3. April 2017
   
4. Take your machine learning skills to the next level by mastering Deep Learning concepts and algorithms using Python.

复制代码

1. Python Deep Learning
   
2. 
   
3. This is the code repository for Python Deep Learning, published by Packt. It contains all the supporting project files necessary to work through the book from start to finish.
   
4. 
   
5. About the Book
   
6. 
   
7. With an increasing interest in AI around the world, deep learning has attracted a great deal of public attention. Every day, deep learning algorithms are used broadly across different industries.
   
8. 
   
9. The book will give you all the practical information available on the subject, including the best practices, using real-world use cases. You will learn to recognize and extract information to increase predictive accuracy and optimize results.
   
10. 
    
11. Starting with a quick recap of important machine learning concepts, the book will delve straight into deep learning principles using Sci-kit learn. Moving ahead, you will learn to use the latest open source libraries such as Theano, Keras, Google's TensorFlow, and H20. Use this guide to uncover the difficulties of pattern recognition, scaling data with greater accuracy and discussing deep learning algorithms and techniques.
    
12. 
    
13. Instructions and Navigation
    
14. 
    
15. All of the code is organized into folders. Each folder starts with a number followed by the application name. For example, Chapter02.
    
16. 
    
17. The code will look like the following:
    
18. 
    
19. mlp.fit(data_train, labels_train)
    
20. pred = mlp.predict(data_test)
    
21. print('Misclassified samples: %d' % (labels_test != pred).sum())
    
22. from sklearn.metrics import accuracy_score print('Accuracy: %.2f' % accuracy_score(labels_test, pred))
    
23. Related Products
    
24. 
    
25. Python: Deeper Insights into Machine Learning
    
26. 
    
27. Deep Learning with Keras
    
28. 
    
29. Deep Learning with Hadoop
    
30. 
    
31. Python Machine Learning

复制代码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

关键词：Learning earning python GitHub Learn

1. import numpy
   
2. from matplotlib.colors import ListedColormap
   
3. import matplotlib.pyplot as plt
   
4. 
   
5. def tanh(x):     
   
6.     return (1.0 - numpy.exp(-2*x))/(1.0 + numpy.exp(-2*x))
   
7. 
   
8. def tanh_derivative(x):     
   
9.     return (1 + tanh(x))*(1 - tanh(x))
   
10.    
    
11. class NeuralNetwork:
    
12.     #network consists of a list of integers, indicating
    
13.     #the number of neurons in each layer
    
14.     def __init__(self, net_arch):
    
15.         numpy.random.seed(0)                  
    
16.         self.activity = tanh         
    
17.         self.activity_derivative = tanh_derivative
    
18.         self.layers = len(net_arch)         
    
19.         self.steps_per_epoch = 1000
    
20.         self.arch = net_arch        
    
21. 
    
22.         self.weights = []         
    
23.         #range of weight values (-1,1)         
    
24.         for layer in range(len(net_arch) - 1):            
    
25.             w = 2*numpy.random.rand(net_arch[layer] + 1, net_arch[layer+1]) - 1           
    
26.             self.weights.append(w)
    
27. 
    
28.     def fit(self, data, labels, learning_rate=0.1, epochs=10):         
    
29.         #Add bias units to the input layer         
    
30.         ones = numpy.ones((1, data.shape[0]))        
    
31.         Z = numpy.concatenate((ones.T, data), axis=1)
    
32.         training = epochs*self.steps_per_epoch
    
33. 
    
34. 
    
35.         for k in range(training):            
    
36.             if k % self.steps_per_epoch == 0:                  
    
37.                 #print ('epochs:', k/self.steps_per_epoch)   
    
38.                 print('epochs: {}'.format(k/self.steps_per_epoch))              
    
39.                 for s in data:                     
    
40.                     print(s, nn.predict(s))
    
41. 
    
42.             sample = numpy.random.randint(data.shape[0])            
    
43.             y = [Z[sample]]
    
44. 
    
45.             for i in range(len(self.weights)-1):                     
    
46.                 activation = numpy.dot(y[i], self.weights[i])                        
    
47.                 activity = self.activity(activation)  
    
48.                 #add the bias for the next layer                     
    
49.                 activity = numpy.concatenate((numpy.ones(1), numpy.array(activity)))                     
    
50.                 y.append(activity)   
    
51.             
    
52.             #last layer              
    
53.             activation = numpy.dot(y[-1], self.weights[-1])            
    
54.             activity = self.activity(activation)            
    
55.             y.append(activity)
    
56.                     
    
57.             #error for the output layer            
    
58.             error = labels[sample] - y[-1]            
    
59.             delta_vec = [error * self.activity_derivative(y[-1])]
    
60. 
    
61.             #we need to begin from the back from the next to last layer
    
62.             for i in range(self.layers-2, 0, -1):  
    
63.                 #delta_vec [1].dot(self.weights[i][1:].T)               
    
64.                 error = delta_vec[-1].dot(self.weights[i][1:].T)
    
65.                 error = error*self.activity_derivative(y[i][1:])               
    
66.                 delta_vec.append(error)
    
67. 
    
68.             # reverse
    
69.             # [level3(output)->level2(hidden)]  => [level2(hidden)->level3(output)]
    
70.             delta_vec.reverse()
    
71. 
    
72.             # backpropagation
    
73.             # 1. Multiply its output delta and input activation
    
74.             #    to get the gradient of the weight.
    
75.             # 2. Subtract a ratio (percentage) of the gradient from the weight
    
76.             for i in range(len(self.weights)):
    
77.                 layer = y[i].reshape(1, nn.arch[i]+1)
    
78. 
    
79.                 delta = delta_vec[i].reshape(1, nn.arch[i+1])
    
80.                 self.weights[i] += learning_rate * layer.T.dot(delta)
    
81. 
    
82.     def predict(self, x):
    
83.         val = numpy.concatenate((numpy.ones(1).T, numpy.array(x)))      
    
84.         for i in range(0, len(self.weights)):
    
85.             val = self.activity(numpy.dot(val, self.weights[i]))
    
86.             val = numpy.concatenate((numpy.ones(1).T, numpy.array(val)))
    
87.             
    
88.         return val[1]
    
89. 
    
90.     def plot_decision_regions(self, X, y, points=200):
    
91.         markers = ('o', '^')
    
92.         colors = ('red', 'blue')
    
93.         cmap = ListedColormap(colors)
    
94.         # plot the decision surface
    
95.         x1_min, x1_max = X[:, 0].min() - 1, X[:, 0].max() + 1
    
96.         x2_min, x2_max = X[:, 1].min() - 1, X[:, 1].max() + 1
    
97.         
    
98.         resolution = max(x1_max - x1_min, x2_max - x2_min)/float(points)
    
99.         #resolution = 0.01
    
100.      
     
101.         xx1, xx2 = numpy.meshgrid(numpy.arange(x1_min, x1_max, resolution), numpy.arange(x2_min, x2_max, resolution))
     
102.         input = numpy.array([xx1.ravel(), xx2.ravel()]).T
     
103.         Z = numpy.empty(0)
     
104.         for i in range(input.shape[0]):
     
105.             val = nn.predict(numpy.array(input[i]))
     
106.             if val < 0.5: val = 0
     
107.             if val >= 0.5: val = 1
     
108.             Z = numpy.append(Z, val)
     
109. 
     
110.         Z = Z.reshape(xx1.shape)
     
111.         
     
112.         plt.pcolormesh(xx1, xx2, Z, cmap=cmap)
     
113.         plt.xlim(xx1.min(), xx1.max())
     
114.         plt.ylim(xx2.min(), xx2.max())
     
115.         # plot all samples
     
116. 
     
117.         classes = ["False", "True"]
     
118.         for idx, cl in enumerate(numpy.unique(y)):
     
119.             plt.scatter(x=X[y == cl, 0], y=X[y == cl, 1], alpha=1.0, c=cmap(idx), marker=markers[idx], s=80, label=classes[idx])
     
120.             
     
121.         plt.xlabel('x-axis')            
     
122.         plt.ylabel('y-axis')
     
123.         plt.legend(loc='upper left')
     
124.         plt.show()            
     
125. 
     
126. if __name__ == '__main__':
     
127. 
     
128.     nn = NeuralNetwork([2,2,1])
     
129. 
     
130.     X = numpy.array([[0, 0],
     
131.                      [0, 1],
     
132.                      [1, 0],
     
133.                      [1, 1]])
     
134. 
     
135.     y = numpy.array([0, 1, 1, 0])
     
136. 
     
137.     nn.fit(X, y, epochs=10)
     
138. 
     
139.     print "Final prediction"
     
140.     for s in X:
     
141.         print(s, nn.predict(s))
     
142.         
     
143.     nn.plot_decision_regions(X, y)

复制代码

1. # -*- coding: utf-8 -*-
   
2. 
   
3. from keras.datasets import mnist
   
4. from keras.models import Sequential
   
5. from keras.layers.core import Dense, Activation
   
6. from keras.utils import np_utils
   
7. 
   
8. (X_train, Y_train), (X_test, Y_test) = mnist.load_data()
   
9. 
   
10. X_train = X_train.reshape(60000, 784)     
    
11. X_test = X_test.reshape(10000, 784)
    
12. 
    
13. classes = 10
    
14. Y_train = np_utils.to_categorical(Y_train, classes)     
    
15. Y_test = np_utils.to_categorical(Y_test, classes)
    
16. 
    
17. input_size = 784
    
18. batch_size = 100     
    
19. hidden_neurons = 100     
    
20. epochs = 30
    
21. 
    
22. model = Sequential()     
    
23. model.add(Dense(hidden_neurons, input_dim=input_size))
    
24. model.add(Activation('sigmoid'))     
    
25. model.add(Dense(classes, input_dim=hidden_neurons))
    
26. model.add(Activation('softmax'))
    
27. 
    
28. model.compile(loss='categorical_crossentropy', metrics=['accuracy'], optimizer='sgd')
    
29. 
    
30. model.fit(X_train, Y_train, batch_size=batch_size, epochs=epochs, verbose=1)
    
31. 
    
32. score = model.evaluate(X_test, Y_test, verbose=1)
    
33. print('Test accuracy:', score[1])
    
34. 
    
35. 
    
36. weights = model.layers[0].get_weights()
    
37. 
    
38. import matplotlib.pyplot as plt     
    
39. import matplotlib.cm as cm
    
40. import numpy
    
41. 
    
42. fig = plt.figure()
    
43.   
    
44. w = weights[0].T         
    
45. for neuron in range(hidden_neurons):         
    
46.     ax = fig.add_subplot(10, 10, neuron+1)
    
47.     ax.axis("off")
    
48.     ax.imshow(numpy.reshape(w[neuron], (28, 28)), cmap = cm.Greys_r)
    
49. 
    
50. plt.savefig("neuron_images.png", dpi=300)   
    
51. plt.show()  

复制代码

1. import tensorflow as tf
   
2. from tensorflow.examples.tutorials.mnist import input_data
   
3. 
   
4. VISIBLE_NODES = 784
   
5. HIDDEN_NODES = 400
   
6. LEARNING_RATE = 0.01
   
7. 
   
8. mnist = input_data.read_data_sets("MNIST_data/")
   
9. 
   
10. input_placeholder = tf.placeholder("float", shape=(None, VISIBLE_NODES))
    
11. 
    
12. weights = tf.Variable(tf.random_normal((VISIBLE_NODES, HIDDEN_NODES), mean=0.0, stddev=1. / VISIBLE_NODES))
    
13. hidden_bias = tf.Variable(tf.zeros([HIDDEN_NODES]))
    
14. visible_bias = tf.Variable(tf.zeros([VISIBLE_NODES]))
    
15. 
    
16. hidden_activation = tf.nn.sigmoid(tf.matmul(input_placeholder, weights) + hidden_bias)
    
17. visible_reconstruction = tf.nn.sigmoid(tf.matmul(hidden_activation, tf.transpose(weights)) + visible_bias)
    
18. 
    
19. final_hidden_activation = tf.nn.sigmoid(tf.matmul(visible_reconstruction, weights) + hidden_bias)
    
20. 
    
21. positive_phase = tf.matmul(tf.transpose(input_placeholder), hidden_activation)
    
22. negative_phase = tf.matmul(tf.transpose(visible_reconstruction), final_hidden_activation)
    
23. 
    
24. weight_update = weights.assign_add(LEARNING_RATE * (positive_phase - negative_phase))
    
25. visible_bias_update = visible_bias.assign_add(LEARNING_RATE *
    
26.                                               tf.reduce_mean(input_placeholder - visible_reconstruction, 0))
    
27. hidden_bias_update = hidden_bias.assign_add(LEARNING_RATE *
    
28.                                             tf.reduce_mean(hidden_activation - final_hidden_activation, 0))
    
29. 
    
30. train_op = tf.group(weight_update, visible_bias_update, hidden_bias_update)
    
31. 
    
32. loss_op = tf.reduce_sum(tf.square(input_placeholder - visible_reconstruction))
    
33. 
    
34. session = tf.Session()
    
35. 
    
36. session.run(tf.initialize_all_variables())
    
37. 
    
38. current_epochs = 0
    
39. 
    
40. for i in range(20):
    
41.     total_loss = 0
    
42.     while mnist.train.epochs_completed == current_epochs:
    
43.         batch_inputs, batch_labels = mnist.train.next_batch(100)
    
44.         _, reconstruction_loss = session.run([train_op, loss_op], feed_dict={input_placeholder: batch_inputs})
    
45.         total_loss += reconstruction_loss
    
46. 
    
47.     print("epochs %s loss %s" % (current_epochs, reconstruction_loss))
    
48.     current_epochs = mnist.train.epochs_completed

复制代码

1. # -*- coding: utf-8 -*-
   
2. 
   
3. import numpy  
   
4. import theano  
   
5. import matplotlib.pyplot as plt
   
6. import theano.tensor as T
   
7. from theano.tensor.nnet import conv
   
8. import skimage.data
   
9. 
   
10. import matplotlib.cm as cm
    
11. 
    
12. depth = 4
    
13. filter_shape = (3, 3)
    
14.    
    
15. input = T.tensor4(name='input')  
    
16.    
    
17. w_shape = (depth, 3, filter_shape[0], filter_shape[1])
    
18. dist = numpy.random.uniform(-0.2, 0.2, size=w_shape)
    
19. W = theano.shared(numpy.asarray(dist, dtype=input.dtype), name = 'W')                             
    
20. conv_output = conv.conv2d(input, W)   
    
21. output = T.nnet.sigmoid(conv_output)
    
22. f = theano.function([input], output)
    
23. 
    
24. astronaut = skimage.data.astronaut()
    
25. img = numpy.asarray(astronaut, dtype='float32') / 255
    
26. filtered_img = f(img.transpose(2, 0, 1).reshape(1, 3, 512, 512))
    
27. 
    
28. 
    
29. plt.axis('off')
    
30. plt.imshow(img)
    
31. plt.show()  
    
32. for img in range(depth):
    
33.   fig = plt.figure()   
    
34.   plt.axis( 'off')   
    
35.   plt.imshow(filtered_img[0, img, :, :, ], cmap = cm.gray)
    
36.   plt.show()
    
37. 
    
38.   filename = "astro" + str(img)
    
39.   fig.savefig(filename, bbox_inches='tight')
    
40.   

复制代码

1. # -*- coding: utf-8 -*-
   
2. 
   
3. import numpy               
   
4. 
   
5. from keras.datasets import cifar10
   
6. from keras.models import Sequential
   
7. from keras.layers.core import Dense, Activation
   
8. from keras.layers import Convolution2D, MaxPooling2D, Flatten
   
9. from keras.layers import Dropout
   
10. from keras.utils import np_utils
    
11. 
    
12. batch_size = 100     
    
13. hidden_neurons = 200
    
14. classes = 10     
    
15. epochs = 20
    
16. 
    
17. (X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
    
18. 
    
19. 
    
20. Y_train = np_utils.to_categorical(Y_train, classes)     
    
21. Y_test = np_utils.to_categorical(Y_test, classes)
    
22. 
    
23. model = Sequential()
    
24. model.add(Convolution2D(32, (3, 3), input_shape=(32, 32, 3)))
    
25. model.add(Activation('relu'))
    
26. model.add(Convolution2D(32, (3, 3)))  
    
27. model.add(Activation('relu'))
    
28. model.add(MaxPooling2D(pool_size=(2, 2)))
    
29. model.add(Dropout(0.25))   
    
30. 
    
31. model.add(Convolution2D(64, (3, 3)))
    
32. model.add(Activation('relu'))     
    
33. model.add(Convolution2D(64, (3, 3)))     
    
34. model.add(Activation('relu'))     
    
35. model.add(MaxPooling2D(pool_size=(2, 2)))     
    
36. model.add(Dropout(0.25))
    
37.                
    
38. model.add(Flatten())
    
39. 
    
40. model.add(Dense(hidden_neurons))
    
41. model.add(Activation('relu'))
    
42. model.add(Dropout(0.5))      
    
43. model.add(Dense(classes))
    
44. model.add(Activation('softmax'))
    
45.      
    
46. 
    
47. model.compile(loss='categorical_crossentropy', metrics=['accuracy'], optimizer='adadelta')
    
48. 
    
49. model.fit(X_train, Y_train, batch_size=batch_size, epochs=epochs, validation_split = 0.1, verbose=1)
    
50. 
    
51. score = model.evaluate(X_test, Y_test, verbose=1)
    
52. print('Test accuracy:', score[1])
    
53. 
    
54. numpy.set_printoptions(threshold='nan')  
    
55. index = 0   
    
56. for layer in model.layers:      
    
57.   filename = "conv_layer_" + str(index)      
    
58.   f1 = open(filename, 'w+')      
    
59.   f1.write(repr(layer.get_weights()))      
    
60.   f1.close()      
    
61.   print (filename + " has been opened and closed")     
    
62.   index = index+1

复制代码

看看     

22222222

谢谢 分享！