根据波士顿房价预测例子改编的tensorflow股票价格预测

注：本文仅作为tensorflow入门学习参考，对于结果请勿较真

导入相应的包

import tensorflow as tf
import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
from sklearn.utils import shuffle

读取股票数据文件

data = pd.read_csv('000001.SZ.CSV',encoding='gbk',usecols = [4,5,6,7,20,21])
data.head(5)

	开盘价(元)	最高价(元)	最低价(元)	收盘价(元)	市盈率	市净率
0	0.5698	0.5698	0.5698	0.5698	33.5310	10.0482
1	0.5670	0.5670	0.5670	0.5670	33.3667	9.9990
2	0.5642	0.5642	0.5642	0.5642	33.2025	9.9498
3	0.5586	0.5586	0.5586	0.5586	32.8740	9.8514
4	0.5559	0.5559	0.5559	0.5559	32.7098	9.8022

将次日的开盘价作为今日要预测的数据y_pred

y_pred = np.zeros(data.shape[0])
for i in range(len(data)-1):
    y_pred[i] = round(data['开盘价(元)'][i+1],4)
data['y_pred'] = y_pred

data.tail(5)

	开盘价(元)	最高价(元)	最低价(元)	收盘价(元)	市盈率	市净率	y_pred
6139	10.47	10.48	10.42	10.46	6.8451	0.9267	10.47
6140	10.47	10.52	10.41	10.50	6.8713	0.9303	10.51
6141	10.51	10.53	10.45	10.51	6.8778	0.9312	10.52
6142	10.52	10.53	10.48	10.52	6.8844	0.9321	10.54
6143	10.54	10.54	10.46	10.50	6.8713	0.9303	0.00

最后一行的数据是没有y_pred的，所以直接把最后一整行都删掉
我们的目标就是将前六列的数据输入到模型中，然后让模型输出一个尽可能接近y_pred的值
也就是用今日的开盘价(元)、最高价(元)、最低价(元)、收盘价(元)、市盈率、市净率来预测明天的开盘价

stockdatas = data.drop(data.index[-1]).values
x_data = stockdatas[:,:6]
y_data = stockdatas[:,6]

# 输出最后一行的值
print(x_data[-1])
print(y_data[-1])

[10.52   10.53   10.48   10.52    6.8844  0.9321]
10.54

归一化训练样本x_data

scaler = MinMaxScaler()
scaler.fit(x_data)
X_data = scaler.transform(x_data)

X_data[-1]

array([0.59144014, 0.59188214, 0.61441079, 0.59880055, 0.02325307,
       0.00445284])

定义模型 y_pred = x * w + b ,注意第一个矩阵的列数要和第二个矩阵的行数相同

x = tf.placeholder(tf.float32, [None,6], name='X')
y = tf.placeholder(tf.float32, [None,1], name='Y')

with tf.name_scope('Model'):
    w = tf.Variable(tf.random_normal([6,1],stddev=0.01),name='W')
    b = tf.Variable(1.0,name='b')
    def model(x,w,b):
        return tf.matmul(x,w) + b
    pred  = model(x,w,b)

定义损失函数和优化器

train_epochs = 3
learning_rate = 0.01
with tf.name_scope('lossfunc'):
    loss_function = tf.reduce_mean(tf.pow(y-pred,2))
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss_function)

开始训练

sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)

for e in range(train_epochs):
    loss_sum = 0.0
    for xs, ys in zip(X_data, y_data):
        xs = xs.reshape(1,6)
        ys = ys.reshape(1,1)
        _,loss = sess.run([optimizer,loss_function],feed_dict={x:xs,y:ys})
        loss_sum = loss_sum + loss
    X_data, y_data = shuffle(X_data,y_data)
    b0 = sess.run(b)
    w0 = sess.run(w)
    loss_avg = loss_sum/len(y_data)
    print('epoch={},lossavg={},w={},b={}'.format(e,loss_avg,w0,b0))

epoch=0,lossavg=0.13093365134630774,w=[[2.2813892]
 [2.410216 ]
 [2.3408403]
 [2.4081025]
 [2.2087398]
 [2.3941853]],b=4.765807628631592
epoch=1,lossavg=0.19175580245865206,w=[[4.057511  ]
 [4.264938  ]
 [4.190086  ]
 [4.2901344 ]
 [0.01645006]
 [0.01836737]],b=0.41750502586364746
epoch=2,lossavg=0.012530410840827653,w=[[4.0354447 ]
 [4.2829742 ]
 [4.2025666 ]
 [4.34924   ]
 [0.01506332]
 [0.01617897]],b=0.4302012026309967

训练完成，输一行数据进去看看效果

xtest = X_data[1050].reshape(1,-1)
pred_val = sess.run(pred, feed_dict={x:xtest})

pred_val

array([[16.004889]], dtype=float32)

看看真实的结果

y_data[1050]

15.51