在《python深度学习》这本书中。
一、21页mnist十分类
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
|
导入数据集 (train_images, train_labels), (test_images, test_labels) = mnist.load_data() 初始数据维度: >>> train_images.shape ( 60000 , 28 , 28 ) >>> len (train_labels) 60000 >>> train_labels array([ 5 , 0 , 4 , ..., 5 , 6 , 8 ], dtype = uint8) 数据预处理: train_images = train_images.reshape(( 60000 , 28 * 28 )) train_images = train_images.astype( 'float32' ) / 255 train_labels = to_categorical(train_labels) 之后: print (train_images, type (train_images), train_images.shape, train_images.dtype) print (train_labels, type (train_labels), train_labels.shape, train_labels.dtype) 结果: [[ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ] ... [ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ]] < class 'numpy.ndarray' > ( 60000 , 784 ) float32 [[ 0. 0. 0. ... 0. 0. 0. ] [ 1. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ] ... [ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 0. 0. ] [ 0. 0. 0. ... 0. 1. 0. ]] < class 'numpy.ndarray' > ( 60000 , 10 ) float32 |
二、51页IMDB二分类
导入数据:
from keras.datasets import imdb (train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words=10000)
参数 num_words=10000 的意思是仅保留训练数据中前 10 000 个最常出现的单词。
train_data和test_data都是numpy.ndarray类型,都是一维的(共25000个元素,相当于25000个list),其中每个list代表一条评论,每个list中的每个元素的值范围在0-9999 ,代表10000个最常见单词的每个单词的索引,每个list长度不一,因为每条评论的长度不一,例如train_data中的list最短的为11,最长的为189。
train_labels和test_labels都是含25000个元素(元素的值要不0或者1,代表两类)的list。
数据预处理:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
|
# 将整数序列编码为二进制矩阵 def vectorize_sequences(sequences, dimension = 10000 ): # Create an all-zero matrix of shape (len(sequences), dimension) results = np.zeros(( len (sequences), dimension)) for i, sequence in enumerate (sequences): results[i, sequence] = 1. # set specific indices of results[i] to 1s return results x_train = vectorize_sequences(train_data) x_test = vectorize_sequences(test_data) 第一种方式:shape为( 25000 ,) y_train = np.asarray(train_labels).astype( 'float32' ) #就用这种方式就行了 y_test = np.asarray(test_labels).astype( 'float32' ) 第二种方式:shape为( 25000 , 1 ) y_train = np.asarray(train_labels).astype( 'float32' ).reshape( 25000 , 1 ) y_test = np.asarray(test_labels).astype( 'float32' ).reshape( 25000 , 1 ) 第三种方式:shape为( 25000 , 2 ) y_train = to_categorical(train_labels) #变成one-hot向量 y_test = to_categorical(test_labels) |
第三种方式,相当于把二分类看成了多分类,所以网络的结构同时需要更改,
最后输出的维度:1->2
最后的激活函数:sigmoid->softmax
损失函数:binary_crossentropy->categorical_crossentropy
预处理之后,train_data和test_data变成了shape为(25000,10000),dtype为float32的ndarray(one-hot向量),train_labels和test_labels变成了shape为(25000,)的一维ndarray,或者(25000,1)的二维ndarray,或者shape为(25000,2)的one-hot向量。
注:
1.sigmoid对应binary_crossentropy,softmax对应categorical_crossentropy
2.网络的所有输入和目标都必须是浮点数张量
补充知识:keras输入数据的方法:model.fit和model.fit_generator
1.第一种,普通的不用数据增强的
1
2
3
4
|
from keras.datasets import mnist,cifar10,cifar100 (X_train, y_train), (X_valid, Y_valid) = cifar10.load_data() model.fit(X_train, Y_train, batch_size = batch_size, nb_epoch = nb_epoch, shuffle = True , verbose = 1 , validation_data = (X_valid, Y_valid), ) |
2.第二种,带数据增强的 ImageDataGenerator,可以旋转角度、平移等操作。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
|
from keras.preprocessing.image import ImageDataGenerator (trainX, trainY), (testX, testY) = cifar100.load_data() trainX = trainX.astype( 'float32' ) testX = testX.astype( 'float32' ) trainX / = 255. testX / = 255. Y_train = np_utils.to_categorical(trainY, nb_classes) Y_test = np_utils.to_categorical(testY, nb_classes) generator = ImageDataGenerator(rotation_range = 15 , width_shift_range = 5. / 32 , height_shift_range = 5. / 32 ) generator.fit(trainX, seed = 0 ) model.fit_generator(generator.flow(trainX, Y_train, batch_size = batch_size), steps_per_epoch = len (trainX) / / batch_size, epochs = nb_epoch, callbacks = callbacks, validation_data = (testX, Y_test), validation_steps = testX.shape[ 0 ] / / batch_size, verbose = 1 ) |
以上这篇keras分类模型中的输入数据与标签的维度实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。
原文链接:https://blog.csdn.net/xiaohuihui1994/article/details/83536752