神经网络（NN）它是一种模仿动物神经网络行为特征，进行分布式并行信息处理的算法数学模型。这种网络依靠系统的复杂程度，通过调整内部大量节点之间相互连接的关系，从而达到处理信息的目的。在提供数据量足够大情况下，神经网络可以拟合出输入到输出之间的任意函数关系。
Tensorflow是一个优秀的深度学习框架，具体有啥好处，可以百度了解哈。
本文分享使用Tensorflow神经网络进行股市的预测
1、数据来源
首先找到一组股票数据，数据可以网络上爬虫，东方财富、大智慧都有。爬虫方法参看以前的文章。
代码语言：javascript复制date = np.linspace(1, 30, 30) #
beginPrice = np.array([2923.19, 2928.06, 2943.92, 2946.26, 2944.40, 2920.85, 2861.33, 2854.58, 2776.69, 2789.02,
2784.18, 2805.59, 2781.98, 2798.05, 2824.49, 2762.34, 2817.57, 2835.52, 2879.08, 2875.47,
2887.66, 2885.15, 2851.02, 2879.52, 2901.63, 2896.00, 2907.38, 2886.94, 2925.94, 2927.75])
endPrice = np.array([2937.36, 2944.54, 2941.01, 2952.34, 2932.51, 2908.77, 2867.84, 2821.50, 2777.56, 2768.68,
2794.55, 2774.75, 2814.99, 2797.26, 2808.91, 2815.80, 2823.82, 2883.10, 2880.00, 2880.33,
2883.44, 2897.43, 2863.57, 2902.19, 2893.76, 2890.92, 2886.24, 2924.11, 2930.15, 2957.41])2、数据展示
基于matplotlib可视化库，建立一个30行2列的矩阵存储股票数据，矩阵的第一列是股票开盘价格，第二列是股票的收盘价格，如果股票的收盘价格高于开盘价格则用红色显示，反之则用绿色显示，可视化股票数据如下图所示。
代码语言：javascript复制for i in range(0, 30): # 画柱状图
dateOne = np.zeros([2])
dateOne[0] = i
dateOne[1] = i
priceOne = np.zeros([2])
priceOne[0] = beginPrice[i]
priceOne[1] = endPrice[i]
if endPrice[i] > beginPrice[i]:
plt.plot(dateOne, priceOne, 'r', lw=6)
else:
plt.plot(dateOne, priceOne, 'g', lw=6)
plt.xlabel("date")
plt.ylabel("price")
plt.show()
3、Tensorflow预测
基于Tensorflow神经网络框架，设计了三层神经网络，其中隐含层包括25个节点，设计的神经网络用来预测股票的收盘价。
代码语言：javascript复制dateNormal = np.zeros([30, 1])
priceNormal = np.zeros([30, 1])
# 归一化
for i in range(0, 30):
dateNormal[i, 0] = i / 29.0
priceNormal[i, 0] = endPrice[i] / 3000.0
x = tf.placeholder(tf.float32, [None, 1])
y = tf.placeholder(tf.float32, [None, 1])
# X->hidden_layer
w1 = tf.Variable(tf.random_uniform([1, 25], 0, 1))
b1 = tf.Variable(tf.zeros([1, 25]))
wb1 = tf.matmul(x, w1) b1
layer1 = tf.nn.relu(wb1) # 激励函数
# hidden_layer->output
w2 = tf.Variable(tf.random_uniform([25, 1], 0, 1))
b2 = tf.Variable(tf.zeros([30, 1]))
wb2 = tf.matmul(layer1, w2) b2
layer2 = tf.nn.relu(wb2)
loss = tf.reduce_mean(tf.square(y - layer2)) # y为真实数据， layer2为网络预测结果
# 梯度下降
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(0, 20000):
sess.run(train_step, feed_dict={x: dateNormal, y: priceNormal})
# 预测， X w1w2 b1b2 -->layer2
pred = sess.run(layer2, feed_dict={x: dateNormal})
date1 = np.linspace(0, 29, 30) #
plt.plot(date1, pred*3000, 'b', lw=3)
plt.show()
运行以上代码可视化神经网络的预测结果如下图所示
完整的代码如下：
代码语言：javascript复制import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
# import tensorflow.compat.v1 as tf
# tf.disable_v2_behavior() # 如果是tensorflow2版本就取消这行注释
date = np.linspace(1, 30, 30) #
beginPrice = np.array([2923.19, 2928.06, 2943.92, 2946.26, 2944.40, 2920.85, 2861.33, 2854.58, 2776.69, 2789.02,
2784.18, 2805.59, 2781.98, 2798.05, 2824.49, 2762.34, 2817.57, 2835.52, 2879.08, 2875.47,
2887.66, 2885.15, 2851.02, 2879.52, 2901.63, 2896.00, 2907.38, 2886.94, 2925.94, 2927.75])
endPrice = np.array([2937.36, 2944.54, 2941.01, 2952.34, 2932.51, 2908.77, 2867.84, 2821.50, 2777.56, 2768.68,
2794.55, 2774.75, 2814.99, 2797.26, 2808.91, 2815.80, 2823.82, 2883.10, 2880.00, 2880.33,
2883.44, 2897.43, 2863.57, 2902.19, 2893.76, 2890.92, 2886.24, 2924.11, 2930.15, 2957.41])
for i in range(0, 30): # 画柱状图
dateOne = np.zeros([2])
dateOne[0] = i
dateOne[1] = i
priceOne = np.zeros([2])
priceOne[0] = beginPrice[i]
priceOne[1] = endPrice[i]
if endPrice[i] > beginPrice[i]:
plt.plot(dateOne, priceOne, 'r', lw=6)
else:
plt.plot(dateOne, priceOne, 'g', lw=6)
plt.xlabel("date")
plt.ylabel("price")
# plt.show()
dateNormal = np.zeros([30, 1])
priceNormal = np.zeros([30, 1])
# 归一化
for i in range(0, 30):
dateNormal[i, 0] = i / 29.0
priceNormal[i, 0] = endPrice[i] / 3000.0
x = tf.placeholder(tf.float32, [None, 1])
y = tf.placeholder(tf.float32, [None, 1])
# X->hidden_layer
w1 = tf.Variable(tf.random_uniform([1, 25], 0, 1))
b1 = tf.Variable(tf.zeros([1, 25]))
wb1 = tf.matmul(x, w1) b1
layer1 = tf.nn.relu(wb1) # 激励函数
# hidden_layer->output
w2 = tf.Variable(tf.random_uniform([25, 1], 0, 1))
b2 = tf.Variable(tf.zeros([30, 1]))
wb2 = tf.matmul(layer1, w2) b2
layer2 = tf.nn.relu(wb2)
loss = tf.reduce_mean(tf.square(y - layer2)) # y为真实数据， layer2为网络预测结果
# 梯度下降
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(0, 20000):
sess.run(train_step, feed_dict={x: dateNormal, y: priceNormal})
# 预测， X w1w2 b1b2 -->layer2
pred = sess.run(layer2, feed_dict={x: dateNormal})
date1 = np.linspace(0, 29, 30) #
plt.plot(date1, pred*3000, 'b', lw=3)
plt.show()代码中需要用到numpy、matplotlib和tensorflow三个库，为了提高下载速度，建议切换到国内的pip源，例如豆瓣、清华等
代码语言：javascript复制pip install numpy -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install matplotlib -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install tensorflow -i https://pypi.tuna.tsinghua.edu.cn/simple
