Troubleshooting my LSTM

[learningpytorch]

Please, I am trying to train a LSTM but I don’t know where it is failing. I have the error:

ModuleAttributeError: ‘LSTM’ object has no attribute ‘hidden_size’.

Full code in Colab: https://colab.research.google.com/drive/1nHODrwukJZYMMnH-fkMTLlvA9o1ihv-r?usp=sharing

``
class LSTM(nn.Module):
def init(self, input_size=1, hidden_size=100):
super().init()
self.hidden= None

    self.lstm1 = nn.LSTM(
        input_size=input_size,
        hidden_size=hidden_size,
        num_layers=1,
        batch_first=True,
        bidirectional=False,
    )
    self.lstm2 = nn.LSTM(
        hidden_size,
        hidden_size * 2,
        num_layers=1,
        batch_first=True,
        bidirectional=False,
    )
    self.fc = nn.Linear(hidden_size * 2, 1)

def forward(self, x):
    x, _ = self.lstm1(x)
    x, _ = self.lstm2(x)

    x = self.fc(x)

    return x

´´

[Usama_Hasan]

Hy, the error is generated at this line.

model.hidden = (torch.zeros(1,1,model.hidden_size).cuda(),
                       torch.zeros(1,1,model.hidden_size).cuda())

Right then you should modify your LSTM class like this.

class LSTM(nn.Module):
   def init(self, input_size=1, hidden_size=100):
        super(LSTM,self).__init__()
        self.hidden_size = hidden_size
        self.hidden= None

Should do

[learningpytorch]

Thanks a million @Usama_Hasan for your help. After changing it, the performance is very poor as you can see in this link:https://colab.research.google.com/drive/1Q7q0dGTylMYMji1E3M104r0149xc9dbh?usp=sharing

I don’t have clear yet the difference between hidden and hidden_size attributes of the model. I am new in this topic, sorry.

Could you recommend me any good tutorial available on the web for this or to improve my knowledge on LSTMs for Time-Series?

Thanks.

[Usama_Hasan]

Sure,
You can check my earlier post on this.

Can LSTM run multivariate time series?

Yes you can use LSTM for time series data prediction. You can find alot of resources for that purpose. You can check this github repo for research papers and link for data resources. https://github.com/Alro10/deep-learning-time-series If you want to checkout for implementation you can also find that in below link. https://github.com/zhangxu0307/time_series_forecasting_pytorch https://github.com/KurochkinAlexey/ConvRNN https://stackoverflow.com/questions/56858924/multivariate-input-lstm-in…

[Usama_Hasan]

First check these on PyTorch website.
https://pytorch.org/tutorials/intermediate/char_rnn_classification_tutorial.html

https://pytorch.org/tutorials/beginner/nlp/sequence_models_tutorial.html

[learningpytorch]

Thanks @Usama_Hasan. The multivariate approach is just what I need. Great info. I will have a look on it today and will get back to you if I have more questions. Thanks again for your support.

[learningpytorch]

Dear @Usama_Hasan, thanks for the information that you provided to me. It is great. However, I still have some questions.

1.- I still do not have clear the differences between hidden and hidden_size in the model.
Complete code in: Google Colab

class LSTM(nn.Module):
    def __init__(self, input_size=1, hidden_size=100):
        super().__init__()
        self.hidden= None
        self.hidden_size=hidden_size

        self.lstm1 = nn.LSTM(
            input_size=input_size,
            hidden_size=hidden_size,
            num_layers=1,
            batch_first=True,
            bidirectional=False,
        )
        self.lstm2 = nn.LSTM(
            hidden_size,
            hidden_size * 2,
            num_layers=1,
            batch_first=True,
            bidirectional=False,
        )
        self.fc = nn.Linear(hidden_size * 2, 1)

    def forward(self, x):
        x, _ = self.lstm1(x)
        x, _ = self.lstm2(x)

        x = self.fc(x)

        return x

2.- Why is the c_n part of the model not used in the second LSTM? Like here
https://colah.github.io/posts/2015-08-Understanding-LSTMs/

Thanks in advance.