keras 的 example 文件 reuters_mlp.py 解析

最新推荐文章于 2022-08-04 16:00:17 发布

zhqh100

最新推荐文章于 2022-08-04 16:00:17 发布

阅读量273

点赞数

分类专栏： TensorFlow python

本文链接：https://blog.youkuaiyun.com/zhqh100/article/details/105363320

版权

python 同时被 2 个专栏收录

51 篇文章

订阅专栏

TensorFlow

48 篇文章

订阅专栏

该文件是一个文本分类的示例，

数据集采用路透社的新闻，我们可以把其内容解析一下：

word_index = reuters.get_word_index()
word_index = {k: (v + 3) for k, v in word_index.items()}
word_index["<PAD>"] = 0
word_index["<START>"] = 1
word_index["?"] = 2  # unknown
word_index["<UNUSED>"] = 3

reverse_word_index = dict([(value, key) for (key, value) in word_index.items()])

def decode_review(text):
    return ' '.join([reverse_word_index.get(i, '?') for i in text])

for i in range(10):
    print(decode_review(x_train[i]))
    print(y_train[i])

就可以看到类似这样的新闻原文：

<START> ? ? said as a result of its december acquisition of ? co it expects earnings per share in 1987 of 1 15 to 1 30 dlrs per share up from 70 cts in 1986 the company said pretax net should rise to nine to 10 mln dlrs from six mln dlrs in 1986 and ? ? revenues to 19 to 22 mln dlrs from 12 5 mln dlrs it said cash ? per share this year should be 2 50 to three dlrs reuter 3

上面的3，是因为 0,1,2,3保留下标，分别表示：“padding,” “start of sequence,“，“unknown.”和“unused”

然后下面采用 Tokenizer 对其进行编码，我们添加一些打印，看下编码前后的关系：

tmp_list = x_train[0]
tmp_list.sort()
print(tmp_list)
print('Vectorizing sequence data...')
tokenizer = Tokenizer(num_words=max_words)
x_train = tokenizer.sequences_to_matrix(x_train, mode='binary')
x_test = tokenizer.sequences_to_matrix(x_test, mode='binary')
print(x_train[0])

打印结果为：

[1, 2, 2, 2, 2, 2, 2, 4, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 8, 8, 8, 9, 10, 11, 11, 11, 11, 12, 15, 15, 15, 15, 15, 15, 16, 16, 17, 19, 19, 22, 22, 22, 25, 26, 29, 30, 32, 39, 43, 44, 48, 48, 49, 52, 67, 67, 67, 83, 84, 89, 90, 90, 90, 102, 109, 111, 124, 132, 134, 151, 154, 155, 186, 197, 207, 209, 209, 258, 270, 272, 369, 447, 482, 504, 864]
Vectorizing sequence data...
[0. 1. 1. 0. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 0. 1. 1. 1. 0. 1. 0. 0. 1. 0.
 0. 1. 1. 0. 0. 1. 1. 0. 1. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 1. 1. 0. 0. 0.
 1. 1. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 0. 0. 0. 0. 1. 1. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 1. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 1. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 1. 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. 0. 0. 0. 1. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 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. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 1. 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. 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.
 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. 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. 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. 0. 0. 0. 0. 0. 0. 0. 0. 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. 0. 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.
 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 0. 0. 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. 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. 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. 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.]

也就是如果文本中包含该字符，则将相应的位置1，否则置0，不过我们也看到，此种编码方式会丢失文本的位置信息；也就是“我要狗”跟“狗咬我”没有区别了，不过好处是节省空间，比起Embedding编码，空间缩小至少一个数量级

编码后的shape为：

x_train shape: (8982, 1000)
y_train shape: (8982, 46)

也就是46种分类，通过一个神经网络来进行分类：

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
dense_1 (Dense)              (None, 512)               512512
_________________________________________________________________
activation_1 (Activation)    (None, 512)               0
_________________________________________________________________
dropout_1 (Dropout)          (None, 512)               0
_________________________________________________________________
dense_2 (Dense)              (None, 46)                23598
_________________________________________________________________
activation_2 (Activation)    (None, 46)                0
=================================================================
Total params: 536,110
Trainable params: 536,110
Non-trainable params: 0
_________________________________________________________________

代码 reuters_mlp_relu_vs_selu.py 使用的是同样的数据集，实现的功能就是比较一下两种神经网络，哪种效果好，也就是激活函数不同，dropout不一样，初始化参数方式有点区别，这里仅打印一下神经网络：

network1

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
dense_1 (Dense)              (None, 16)                16016
_________________________________________________________________
activation_1 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_1 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_2 (Dense)              (None, 16)                272
_________________________________________________________________
activation_2 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_2 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_3 (Dense)              (None, 16)                272
_________________________________________________________________
activation_3 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_3 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_4 (Dense)              (None, 16)                272
_________________________________________________________________
activation_4 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_4 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_5 (Dense)              (None, 16)                272
_________________________________________________________________
activation_5 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_5 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_6 (Dense)              (None, 16)                272
_________________________________________________________________
activation_6 (Activation)    (None, 16)                0
_________________________________________________________________
dropout_6 (Dropout)          (None, 16)                0
_________________________________________________________________
dense_7 (Dense)              (None, 46)                782
_________________________________________________________________
activation_7 (Activation)    (None, 46)                0
=================================================================
Total params: 18,158
Trainable params: 18,158
Non-trainable params: 0
_________________________________________________________________

network2

________________________________________________________________________________
Layer (type)                        Output Shape                    Param #
================================================================================
dense_1 (Dense)                     (None, 16)                      16016
________________________________________________________________________________
activation_1 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_1 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_2 (Dense)                     (None, 16)                      272
________________________________________________________________________________
activation_2 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_2 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_3 (Dense)                     (None, 16)                      272
________________________________________________________________________________
activation_3 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_3 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_4 (Dense)                     (None, 16)                      272
________________________________________________________________________________
activation_4 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_4 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_5 (Dense)                     (None, 16)                      272
________________________________________________________________________________
activation_5 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_5 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_6 (Dense)                     (None, 16)                      272
________________________________________________________________________________
activation_6 (Activation)           (None, 16)                      0
________________________________________________________________________________
alpha_dropout_6 (AlphaDropout)      (None, 16)                      0
________________________________________________________________________________
dense_7 (Dense)                     (None, 46)                      782
________________________________________________________________________________
activation_7 (Activation)           (None, 46)                      0
================================================================================
Total params: 18,158
Trainable params: 18,158
Non-trainable params: 0
________________________________________________________________________________