hichenway / stock_predict_with_lstm

Predict stock with LSTM supporting pytorch, keras and tensorflow

Predict stock with LSTM

This project includes training and predicting processes with LSTM for stock data. The characteristics is as fellow:

• Concise and modular
• Support three mainstream deep learning frameworks of pytorch, keras and tensorflow
• Parameters, models and frameworks can be highly customized and modified
• Supports incremental training
• Support predicting multiple indicators at the same time
• Support predicting any number of days
• Support train visualization and log record

Chinese introduction can refer to : https://blog.csdn.net/songyunli1111/article/details/78513811

The simultaneous predict results for stock high and low price with pytorch show as follow:

stock_predict_with_lstm's Issues

iteration设置大于1的话，没有把上一次的final_state作为下一次的init_state，这样没问题么

这样训练的话每个epoch就是单独训练了吧？望解答，谢谢！

do_train_visualized = False

我把这个do_train_visualized = False改成True会报错：ConnectionRefusedError: [WinError 10061] 由于目标计算机积极拒绝，无法连接。该怎么解决。

dropout并没有使用

stock_predict_2.py 中，定义了drop = tf.nn.rnn_cell.DropoutWrapper(basicLstm, output_keep_prob=keep_prob) ，但是实际并没有使用它，这里产生作用了吗？

会不会有未来函数呢

self.mean = np.mean(self.data, axis=0) # 数据的均值和方差
self.std = np.std(self.data, axis=0)
self.norm_data = (self.data - self.mean)/self.std # 归一化，去量纲

The codes are aimed at Singlevariate time-series or Multivariate time-series?

Excuse me ? I found your datas are all singe variable, however, your paper said that your work is aimed at multivariate time-series, so do you have another codes or if it is my fault to understand this code?

您好，代码运行后有个问题想请教您

ValueError: Attempt to have a second RNNCell use the weights of a variable scope that already has weights: 'sec_lstm/rnn/basic_lstm_cell'; and the cell was not constructed as BasicLSTMCell(..., reuse=True). To share the weights of an RNNCell, simply reuse it in your second calculation, or create a new one with the argument reuse=True.

最近初学tensorflow，在博客上看到您的代码下载了运行后报了错，我在网上搜了很多方案都不行，非常困惑，恳请您能指点一下，谢谢

时间特征为什么没有考虑

作者您好 你的代码中好像没有考虑时间特征

你好，输出图我看不太懂

你好，输出的图我不太明白
plt.plot(list(range(len(normalize_data), len(normalize_data) + len(predict))), predict, color='r')
红色线表示normalize_data之后的预测值，对吗？但是我的形状有些奇怪，没看懂··

非常感谢！

请教关于训练和预测时的数据格式的问题

代码中的注释有写到，在非连续训练模式下，每time_step行数据会作为一个样本，两个样本错开一行，比如：1-20行，2-21行
比如有1000行的数据，n_past为10，特征是5个，那么构建出来的train_x shape为(991,10,5)，train_y shape为(991,10,1)，也就是对于每一个样本，使用过去10条数据来预测结果，并和真值相比较。

但是如果是用来预测，构建的test_X的shape是(100,10,5)，test_Y的shape是（1000,1,5），在源码处也可以看到不是使用连续方式来构建的test_X。
所以构建出来的LSTM模型并不是输入过去n_past条数据，模型输出预测的这一条数据吗？

question about do_continue_train

Hello, thanks for you work.
In your case, while set do_continue_train is False, the train data must be a continuous series and can only be a series. So, my problem is, how should I train my data with serveral multi-series data?
Looking forward to your reply, thank you.

请问如何获取预测后的数据

非常感谢楼主，您的代码直接就能运行了，质量真棒！
想请教个问题：为什么stock_predict_1.py每次执行预测的结果都不一样呢？model_save1里的文件即使删除后重新运行也是不一样的结果。

现象：每次执行计算的next_seq值都不一样，所以预测的结果也不同
next_seq=sess.run(pred,feed_dict={X:[prev_seq]})
print ("------------->"+str(next_seq))