keras分类模型中的输入数据与标签的维度

在《python深度学习》这本书中。
一、21页mnist十分类

导入数据集
from keras.datasets import mnist
(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.]]  (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.]]  (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:

# 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

你可能感兴趣的:(Keras)