深度学习 tensorflow例子

问题：1：构成的网络结构？

         2：输入数据大小？

https://yq.aliyun.com/articles/118726?spm=5176.100239.blogcont122848.14.zwhUbk

import tensorflow as tf
import pandas as pd
import numpy as np
import os
import matplotlib
import matplotlib.pyplot as plt
import random
#matplotlib inline
import tensorflow as tf
import shutil
import tensorflow.contrib.learn as tflearn
import tensorflow.contrib.layers as tflayers
from tensorflow.contrib.learn.python.learn import learn_runner
import tensorflow.contrib.metrics as metrics
import tensorflow.contrib.rnn as rnn


random.seed(111)
rng=pd.date_range(start='2000',periods=209,freq='M')
"""pd.date_range 函数:生成日期范围 
   start：开始日期 periods:日期范围内日期的个数，periods*freq 等于日期范围的长度
   freq：每多少天或其他明确时间频率的方式，默认为D，即1天 M为一个月
"""
ts=pd.Series(np.random.uniform(-10,10,size=len(rng)).cumsum())
ts.plot(c='b',title='Example Time Series')     #颜色蓝色
plt.show()
ts.head(10)                 #使用head查看前几行数据（默认是前5行），不过你可以指定前几行


TS=np.array(ts)
num_periods=20              #观测值
f_horizon=1     #forecast horizon 预测一步

x_data=TS[:(len(TS)-len(TS)%num_periods)]          #len(TS)=209
x_batches=x_data.reshape(-1,20,1)

y_data=TS[1:(len(TS)-(len(TS)%num_periods))+f_horizon]
y_batches=y_data.reshape(-1,20,1)

print("x_data shape%s"%len(TS))
print("x_batchesis%s"%len(x_batches))             #10
print(x_batches.shape)                    # (10, 20, 1)
print("x_batches[0:2]is %s"%x_batches[0:2])


print("y_batches[0:1]is%s"%y_batches[0:1])
print(y_batches.shape)                      #(10, 20, 1)

##??不懂 测试数据是怎么取的 为什么
def test_data(series,forecast,num_periods):
    test_x_setup=TS[-(num_periods+forecast):]
    testX=test_x_setup[:num_periods].reshape(-1,20,1)
    testY=TS[-(num_periods):].reshape(-1,20,1)
    return testX,testY

X_test,Y_test=test_data(TS,f_horizon,num_periods)
print(X_test.shape)

print(X_test)


tf.reset_default_graph()              #清除默认图的堆栈，并设置全局图为默认图

num_periods=20
inputs=1
hidden=100
output=1

X=tf.placeholder(tf.float32,[None,num_periods,inputs])
y=tf.placeholder(tf.float32,[None,num_periods,output])

basic_cell=tf.contrib.rnn.BasicRNNCell(num_units=hidden,activation=tf.nn.relu)
rnn_output,states=tf.nn.dynamic_rnn(basic_cell,X,dtype=tf.float32)

learning_rate=0.001


"tf.layers.dense:全连接层 操作 outputs = activation(inputs.kernel + bias) "
stacked_rnn_output=tf.reshape(rnn_output,[-1,hidden])            #转为二维数据
stacked_outputs=tf.layers.dense(stacked_rnn_output,output)
outputs=tf.reshape(stacked_outputs,[-1,num_periods,output])              #?

loss=tf.reduce_sum(tf.square(outputs-y))
optimizer=tf.train.AdamOptimizer(learning_rate=learning_rate)
training_op=optimizer.minimize(loss)

init=tf.global_variables_initializer()

####################################
epochs=1000
with tf.Session() as sess:
    init.run()
    for ep in range(epochs):
        sess.run(training_op,feed_dict={X:x_batches,y:y_batches})
        if ep %100 ==0:
            mse=loss.eval(feed_dict={X:x_batches,y:y_batches})
            print(ep,"\TMSE:",mse)

    y_pred=sess.run(outputs,feed_dict={X:X_test})
    print(y_pred)

plt.title("Forecast vs Actual",fontsize=14)
plt.plot(pd.Series(np.ravel(Y_test)),"bo",markersize=10,label="Actual")
#plt.plot(pd.Series(np.ravel(Y_test)),"w*",markersize=10)
plt.plot(pd.Series(np.ravel(y_pred)),"r.",markersize=10,label="Forecast")
plt.legend(loc="upper left")
plt.xlabel("Time Periods")

plt.show()