详细解读简单的lstm的实例

本文是初学keras这两天来，自己仿照addition_rnn.py，写的一个实例，数据处理稍微有些不同，但是准确性相比addition_rnn.py 差一点，下面直接贴代码，
解释和注释都在代码里边。

<span style="font-family: Arial, Helvetica, sans-serif;">#coding:utf-8</span>

from keras.models import  Sequential
from keras.layers.recurrent import LSTM
from utils import  log
from numpy import random
import numpy as np
from  keras.layers.core import RepeatVector, TimeDistributedDense, Activation

'''
先用lstm实现一个计算加法的keras版本, 根据addition_rnn.py改写
size: 500
10次: test_acu = 0.3050  base_acu= 0.3600
30次: rest_acu = 0.3300  base_acu= 0.4250
size: 50000
10次: test_acu: loss: 0.4749 - acc: 0.8502 - val_loss: 0.4601 - val_acc: 0.8539
      base_acu: loss: 0.3707 - acc: 0.9008 - val_loss: 0.3327 - val_acc: 0.9135
20次: test_acu: loss: 0.1536 - acc: 0.9505 - val_loss: 0.1314 - val_acc: 0.9584
      base_acu: loss: 0.0538 - acc: 0.9891 - val_loss: 0.0454 - val_acc: 0.9919
30次: test_acu: loss: 0.0671 - acc: 0.9809 - val_loss: 0.0728 - val_acc: 0.9766
      base_acu: loss: 0.0139 - acc: 0.9980 - val_loss: 0.0502 - val_acc: 0.9839
'''

log = log()
#defination the global variable
training_size = 50000
hidden_size = 128
batch_size = 128
layers = 1

maxlen = 7
single_digit = 3


def generate_data():
    log.info("generate the questions and answers")
    questions = []
    expected =