Keras II classification problem

Keras Second, classification problem

One 、Keras Development Overview

Keras The model mainly includes the following steps ：
(1) Define training data ： Input tensor and target tensor .
(2) Define the composition network of the layer （ The model ）, Map input to target .
(3) Configure learning process ： Choose the loss function 、 Optimizer and indicators to monitor .
(4) Calling the model fit Methods we iterated on the training data .

Two 、 The classification of film reviews （ Dichotomous problem ）

1. Data sets ：IMDB Data sets （ For an introduction to this dataset, see here ）
2. Hot coding alone (one-hot code ) For an introduction to this code, see REF1、REF2、REF3、REF4.
3.one-hot For the purpose of coding, see The only hot coding in data preprocessing （One-Hot）： Why use one-hot code ？
The following text is excerpted from The only hot coding in data preprocessing （One-Hot）： Why use one-hot code ？: Use one-hot The direct reason for this is that nowadays there are many classifications cnn The output of the network is generally softmax layer , And its output is a probability distribution , Thus, the labels required to be input also appear in the form of probability distribution , Then calculate the cross entropy and so on .one-hot In fact, it gives the real probability distribution of the real sample , The probability of one sample data is 1, The rest are all 0.. When calculating the loss cross entropy , Direct use 1*log（1/ probability ）, The cross entropy is directly calculated , As a loss .

3、 ... and 、 The code is as follows