tensorflow读书报告 

 1 # -*- coding: utf-8 -*-
 2 """
 3 Created on Mon Apr 11 19:10:39 2022
 4  
 5 @author: 10320
 6 """
 7  
 8 import tensorflow as tf
 9 from tensorflow import keras
10  
11 import numpy as np
12 import matplotlib.pyplot as plt
13  
14 fashion_mnist = keras.datasets.fashion_mnist
15  
16 (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
17  
18  
19 class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
20                'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
21 train_images.shape
22 len(train_labels)
23 train_labels
24 test_images.shape
25 len(test_labels)
26  
27 plt.figure()
28 plt.imshow(train_images[0])
29 plt.colorbar()
30 plt.grid(False)
31 plt.show()
32 train_images = train_images / 255.0
33  
34 test_images = test_images / 255.0
35  
36 plt.figure(figsize=(10,10))
37 for i in range(25):
38     plt.subplot(5,5,i+1)
39     plt.xticks([])
40     plt.yticks([])
41     plt.grid(False)
42     plt.imshow(train_images[i], cmap=plt.cm.binary)
43     plt.xlabel(class_names[train_labels[i]])
44 plt.show()
45  
46 model = keras.Sequential([
47     keras.layers.Flatten(input_shape=(28, 28)),
48     keras.layers.Dense(128, activation='relu'),
49     keras.layers.Dense(10)
50 ])
51  
52 model.compile(optimizer='adam',
53               loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
54               metrics=['accuracy'])
55  
56 model.fit(train_images, train_labels, epochs=10)

 

 

 

 

简答题

（1）Tensorflow和Pytorch

（2）可以直接赋值，也可以使用初始化函数

import tensorflow as tf

bias1=tf.Variable(2)

bias2=tf.Variable(initial_value=3.)

 

其他更加复杂的初始化方法 如：tf.zeros\tf.zeros_like\tf.ones_like\tf.random.truncated_normal等等

其中tf.random.truncated_normal和tf.zeros是常常用来进行权值和偏置的初始化方法

 

（3）序贯式和函数式

 1 #序贯式1
 2 import tensorflow as tf
 3 
 4 model = tf.keras.Sequential()
 5 #创建一个全连接层，神经元个数为256，输入为784，激活函数为relu
 6 model.add(tf.keras.layers.Dense(256, activation='relu', input_dim=784))
 7 model.add(tf.keras.layers.Dense(128, activation='relu'))
 8 model.add(tf.keras.layers.Dense(10, activation='softmax'))
 9 
10 #序贯式2
11 import tensorflow as tf
12 
13 imput_layer = tf.keras.layers.Input(shape=(784,))
14 hid1_layer = tf.keras.layers.Dense(256, activation='relu')
15 hid2_layer = tf.keras.layers.Dense(128, activation='relu')
16 output_layers = tf.keras.layers.Dense(10, activation='softmax') #将层的列表传给Sequential的构造函数
17 model = tf.keras.Sequential(layers=[imput_layer, hid1_layer, hid2_layer, output_layers])
18 
19  
20 
21 #函数式
22 import tensorflow as tf
23 #创建一个模型，包含一个输入层和三个全连接层
24 inputs = tf.keras.layers.Input(shape=(4))
25 x=tf.keras.layers.Dense(32,activation='relu')(inputs)
26 x=tf.keras.layers.Dense(64,activation='relu')(x)
27 outputs=tf.keras.layers.Dense(3,activation='softmax')(x)
28 model=tf.keras.Model(inputs=inputs,outputs =outputs)
（4） 
import torch
data=torch.rand(5,3)
print(data)

 

（5）Keras,Caffe,MXNet,Sonnet,Deeplearning4j

posted @ 2022-04-25 00:02  -tsir-  阅读(81)  评论(0)    收藏  举报