ysn2233 / attentioned-dual-stage-stock-prediction

By PyTorch

DA-LSTM

This is an implementation of paper "A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction". I only did a test to predict the price of AAPL.US by its historical data as well as the price of its opponent MSFT.US.

Dataset

Downloaded from NASDAQ 100 STOCK DATA.

Argument

-e, --epoch - the number of epochs

-b, --batch - the batch size

-s, --split - the split ratio of train and test set

-i, --interval - save models every interval epochs

-l, --lrate - learning rate of optimizor

-t, —test - test phase

-m, —model - if in test phase, the models name(if model name is "encoder50" and decoder50", inptut 50)

Sample train

Traing 500 epochs, with batch-size 128, save models every 100 epochs.

Python3 trainer -e 500 -b 128 -i 100

Sample test

Test data use model "encoder50" and "decoder50"

Python3 trainer -t -m 50

attentioned-dual-stage-stock-prediction's Issues

How can I solve "RuntimeError:"?

When I ran the code in test phase, I got this error

Traceback (most recent call last):
File "trainer.py", line 152, in
trainer.test(mname, batch_size)
File "trainer.py", line 66, in test
y_pred_train = self.predict(x_train, y_train, y_seq_train, batch_size)
File "trainer.py", line 90, in predict
y_res = self.decoder(code, var_y_input)
File "/home/tatsuhiro/PycharmProjects/untitled/venv/lib/python3.5/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/tatsuhiro/PycharmProjects/untitled/attentioned-dual-stage-stock-prediction-master/model.py", line 101, in forward
ct = torch.bmm(beta_t.unsqueeze(1), h)
RuntimeError: invalid argument 6: wrong matrix size at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:477

I do not change any of programs.
How can I solve this problem?

模型不易扩展

这种写法，只适合外部驱动序列只有一个的情况，如果多个，模型要修改的地方很多，不适合扩展性。

Attention求教

老哥，能不能在代码中对照着论文大概解释一下两个阶段的Attention啊？

torch.bmm(bt, h) 这里会出现维度不匹配而出错，请问怎么解决呢

        bt = beta_t.unsqueeze(1)
        ct = torch.bmm(bt, h)