使用TensorFlow2.0实现一个简单的股票预测

使用TensorFlow2.0实现一个简单的股票预测

1、环境

Python3.7.4
PyCharm 2019.1.3 (Professional Edition)
Windows 10
Tensorflow2.0+cuda10.0

2、数据集及预处理

• 数据集格式为一维数据，每行代表一天的数据，将数据集每十一天划分为一组，前十天为训练数据，第十一天作为标签数据，如果大于第十天数据，则为1，否则为0。即问题是一个二分类任务。根据前十天走势判断第十一天上涨还是下跌。
• 预处理过程将数据集每十一天作为一组数据，其中前十天是训练数据，第十一天是标签。数据集取csv文件中前119702行数据（刚好被11整除），119702/11即10882组数据，取百分之八十作为训练集（8705组），百分之二十作为测试集（2177组）。 则数据格式分为训练集样本[8705,10,1],训练集标签[8705,1]，测试集样本[2177,10,1]，测试集标签[2177,1]。具体过程见代码。

3、源代码

import os
import tensorflow as tf
import numpy as np
import csv
import copy
from tensorflow import keras
from tensorflow.keras import layers

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
assert tf.__version__.startswith('2.')

def preprocessing():
    data = csv.reader(open('股票数据.csv', encoding='ANSI'))
    count = 0 #划分训练集和测试集，80%作为训练集
    group = 11 #每十一个数据作为一组
    train_x = []
    train_y = []
    test_x = []
    test_y = []
    ten = 0
    group_list = []
    line_list = []
    for line in data:
        if group == 1:   #第十一天为标签数据
            if float(line[0]) > ten:   #第十一天大于第十天为正，即1
                list = [1]
                if count > 95755:
                    test_y.append(copy.deepcopy(list))
                else:
                    train_y.append(copy.deepcopy(list))
            else:
                list = [0]
                if count > 95755:
                    test_y.append(copy.deepcopy(list))
                else:
                    train_y.append(copy.deepcopy(list))
            group = 11
            if count > 95755:
                test_x.append(copy.deepcopy(group_list))      #使用深拷贝
            else:
                train_x.append(copy.deepcopy(group_list))
            group_list.clear()
        else:
            if group == 2:   #保存第十天的数据，用于和第十一天比较
                ten = float(line[0])
            line_list.append(float(line[0]))
            #group_list.append(float(line[0]))
            group_list.append(copy.deepcopy(line_list))
            line_list.clear()
            group = group - 1
        count = count + 1

    print(tf.constant(train_x))   #[8705,10,1]
    print(tf.constant(train_y))   #[8705,1]
    print(tf.constant(test_x))    #[2177,10,1]
    print(tf.constant(test_y))    #[2177,1]
    return tf.constant(train_x),tf.constant(train_y),tf.constant(test_x),tf.constant(test_y)

class MyRNN(keras.Model):

    def __init__(self, units):
        super(MyRNN, self).__init__()

        self.rnn = keras.Sequential([
            layers.LSTM(units, dropout=0.5, return_sequences=True),
            layers.LSTM(units, dropout=0.5)
        ]
        )

        self.outlayer = layers.Dense(1)

    def call(self, inputs, training=None):
        x = inputs
        x = self.rnn(x)
        x = self.outlayer(x)
        prob = tf.sigmoid(x)
        return prob


def main():
    train_x,train_y,test_x,test_y = preprocessing()
    batchsz = 128

    db_train = tf.data.Dataset.from_tensor_slices((train_x, train_y))
    db_train = db_train.shuffle(1000).batch(batchsz, drop_remainder=True)  # 可能不是batchsz的整数倍，将最后一部分丢弃掉
    db_test = tf.data.Dataset.from_tensor_slices((test_x, test_y))
    db_test = db_test.batch(batchsz, drop_remainder=True)

    print(db_train)
    print(db_test)

    units = 64
    epochs = 40

    model = MyRNN(units)
    model.compile(optimizer = keras.optimizers.Adam(0.001),
                  loss = tf.losses.BinaryCrossentropy(),
                  metrics=['accuracy'])
    model.fit(db_train, epochs=epochs, validation_data=db_test)
    model.evaluate(db_test)

if __name__ == '__main__':
    main()