第二章的代码

Author: sunjiaxin111

.gitignore

@@ -99,3 +99,4 @@ ENV/
 
 # mypy
 .mypy_cache/
+.idea*

chapter2/a-first-look-at-a-neural-network.py

@@ -0,0 +1,45 @@
+from keras.datasets import mnist
+from keras import models
+from keras import layers
+from keras.utils import to_categorical
+
+# 读取Keras自带的mnist数据集
+(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
+
+# 查看训练数据
+print(train_images.shape)
+print(len(train_labels))
+print(train_labels)
+
+# 查看测试数据
+print(test_images.shape)
+print(len(test_labels))
+print(test_labels)
+
+# 网络架构
+network = models.Sequential()
+network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))
+network.add(layers.Dense(10, activation='softmax'))
+
+# 编译
+network.compile(optimizer='rmsprop',
+                loss='categorical_crossentropy',
+                metrics=['accuracy'])
+
+# 预处理图像数据
+train_images = train_images.reshape((60000, 28 * 28))
+train_images = train_images.astype('float32') / 255
+
+test_images = test_images.reshape((10000, 28 * 28))
+test_images = test_images.astype('float32') / 255
+
+# 预处理图像标签
+train_labels = to_categorical(train_labels)
+test_labels = to_categorical(test_labels)
+
+# 训练网络
+network.fit(train_images, train_labels, epochs=5, batch_size=128)
+
+# 评价网络
+test_loss, test_acc = network.evaluate(test_images, test_labels)
+print('test_acc:', test_acc)

chapter2/tensor-operation.py

@@ -0,0 +1,136 @@
+import numpy as np
+
+
+# 基于元素的relu原生实现
+def naive_relu(x):
+    # x是一个二维张量
+    assert len(x.shape) == 2
+
+    x = x.copy()  # 避免覆盖输入的张量
+    for i in range(x.shape[0]):
+        for j in range(x.shape[1]):
+            x[i, j] = max(x[i, j], 0)
+    return x
+
+
+# 基于元素的add原生实现
+def naive_add(x, y):
+    # x和y是二维张量
+    assert len(x.shape) == 2
+    assert x.shape == y.shape
+
+    x = x.copy()  # 避免覆盖输入的张量
+    for i in range(x.shape[0]):
+        for j in range(x.shape[1]):
+            x[i, j] += y[i, j]
+    return x
+
+
+# 只能广播最后一个维度、最后两个维度。。。
+x = np.array([[[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]]])
+# y = np.array([1, 2, 3])  # 维度不匹配
+# y = np.array([1, 2, 3, 4])  # 维度不匹配
+# y = np.array([1, 2, 3, 4, 5])
+# y = np.array([[1, 2, 3, 4, 5],
+#               [2, 3, 4, 5, 6],
+#               [1, 2, 3, 4, 5],
+#               [1, 2, 3, 4, 5]])  # 维度不匹配
+y = np.array([[1, 2, 3, 4, 5],
+              [2, 3, 4, 5, 6],
+              [1, 2, 3, 4, 5]])
+z = x + y
+print(z)
+
+
+# 矩阵加向量的原生实现
+def naive_add_matrix_and_vector(x, y):
+    # x是一个二维张量
+    # y是一个向量
+    assert len(x.shape) == 2
+    assert len(y.shape) == 1
+    assert x.shape[1] == y.shape[0]
+
+    x = x.copy()  # 避免覆盖输入的张量
+    for i in range(x.shape[0]):
+        for j in range(x.shape[1]):
+            x[i, j] += y[j]
+    return x
+
+
+# x是一个随机张量，大小为(64, 3, 32, 10)
+x = np.random.random((64, 3, 32, 10))
+# y是一个随机张量，大小为(32, 10)
+y = np.random.random((32, 10))
+
+# z的大小为(64, 3, 32, 10)
+z = np.maximum(x, y)
+print(z.shape)
+
+
+# 向量dot的原生实现
+def naive_vector_dot(x, y):
+    # x和y是向量
+    assert len(x.shape) == 1
+    assert len(y.shape) == 1
+    assert x.shape[0] == y.shape[0]
+
+    z = 0
+    for i in range(x.shape[0]):
+        z += x[i] * y[i]
+    return z
+
+
+# 矩阵-向量dot的原生实现
+def naive_matrix_vector_dot(x, y):
+    # x是矩阵
+    # y是向量
+    assert len(x.shape) == 2
+    assert len(y.shape) == 1
+    assert x.shape[1] == y.shape[0]
+
+    z = np.zeros(x.shape[0])
+    for i in range(x.shape[0]):
+        # for j in range(x.shape[1]):
+        #     z[i] += x[i][j] * y[j]
+        z[i] = naive_vector_dot(x[i, :], y)
+    return z
+
+
+x = np.array([[1, 2, 3],
+              [1, 2, 3]])
+y = np.array([1, 2, 3])
+z = naive_matrix_vector_dot(x, y)
+print(z)
+
+
+# 矩阵dot的原生实现
+def naive_matrix_dot(x, y):
+    # x和y是矩阵
+    assert len(x.shape) == 2
+    assert len(y.shape) == 2
+    assert x.shape[1] == y.shape[0]
+
+    z = np.zeros((x.shape[0], y.shape[1]))
+    for i in range(y.shape[1]):
+        z[:, i] = naive_matrix_vector_dot(x, y[:, i])
+    return z
+
+
+x = np.array([[1, 2, 3],
+              [1, 2, 3]])
+y = np.array([[1, 2, 3, 4],
+              [1, 2, 3, 4],
+              [1, 2, 3, 4]])
+z = naive_matrix_dot(x, y)
+print(z)

chapter2/tensor-reshaping.py

@@ -0,0 +1,33 @@
+from keras.datasets import mnist
+import numpy as np
+
+(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
+
+# 改形
+train_images = train_images.reshape((60000, 28 * 28))
+
+x = np.array([[0., 1.],
+              [2., 3.],
+              [4., 5.]])
+print(x.shape)
+x = x.reshape((6, 1))
+print(x)
+x = x.reshape((2, 3))
+print(x)
+
+# 矩阵转置
+x = np.zeros((300, 20))
+print(x.shape)
+x = np.transpose(x)
+print(x.shape)
+
+'''
+past_velocity = 0
+momentum = 0.1  # 一个常量因子
+while loss > 0.01:  # 优化循环
+    w, loss, gradient = get_current_parameters()
+    velocity = past_velocity * momentum - learning_rate * gradient  # 书上写的是加号，我认为是减号，看了keras的源码发现确实是减号
+    w = w + momentum * velocity - learning_rate * gradient
+    past_velocity = velocity
+    update_parameter(w)
+'''

chapter2/tensor.py

@@ -0,0 +1,76 @@
+from keras.datasets import mnist
+import numpy as np
+import matplotlib.pyplot as plt
+
+# 标量
+x = np.array(12)
+print(x)
+print(x.shape)
+print(x.ndim)
+
+# 向量
+x = np.array([12, 3, 6, 14])
+print(x)
+print(x.shape)
+print(x.ndim)
+
+# 矩阵
+x = np.array([[5, 78, 2, 34, 0],
+              [6, 79, 3, 35, 1],
+              [7, 80, 4, 36, 2]])
+print(x.shape)
+print(x.ndim)
+
+# 3维张量及高维张量
+x = np.array([[[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]],
+              [[5, 78, 2, 34, 0],
+               [6, 79, 3, 35, 1],
+               [7, 80, 4, 36, 2]]])
+print(x.shape)
+print(x.ndim)
+
+# 查看MNIST数据集的维度和尺寸
+(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
+print(train_images.ndim)
+print(train_images.shape)
+print(train_images.dtype)
+
+# 画出训练集中的一张图片
+digit = train_images[4]
+
+plt.imshow(digit, cmap=plt.cm.binary)
+plt.show()
+
+# tensor切片
+my_slice = train_images[10:100]
+print(my_slice.shape)
+
+# 等价切片1
+my_slice = train_images[10:100, :, :]
+print(my_slice.shape)
+
+# 等价切片2
+my_slice = train_images[10:100, 0:28, 0:28]
+print(my_slice.shape)
+
+# 切右下角的14*14像素
+my_slice = train_images[:, 14:, 14:]
+
+# 切正中间的14*14像素
+my_slice = train_images[:, 7:-7, 7:-7]
+
+# 批量数据
+batch = train_images[:128]
+# 下一批数据
+batch = train_images[128:256]
+# 第n个批量
+n = 2
+batch = train_images[128 * n:128 * (n + 1)]