patrickloeber / pytorch-examples Goto Github PK

Hello @python-engineer ,

thank you for the amazing work. I am trying to implement the LSTM architecture to my custom time series data, I want to make sure i have the best possible architecture of the LSTM, can you advise on what parameters shall i keep in mind to finalize a model, except for hyperparameters. I will keep testing the hyperparameters.

The dataset has around 300k timestamps.
Thanks in advance.

Hi,

In the LSTM code example, you use both lower an upper case n, I struggle to distinguish the difference between the two or if they are both referring to batch sizes?
I'm utilise this code to learn how to implement my own LSTM model. but I am getting an error
"RuntimeError: Expected hidden size (1, 4, 128), got [1, 28, 128]"
In this instance my batch size is 4, and number of layers is one, hense the 2nd array/tensor size is [num layers, image size, number of hidden layers].

The expected hidden size appears to be [num layers, batch size, num hidden layers]
I suspect I'll debug this soon, but I think that the following lines could be clearer in your code if N was defined in the comment.

you and your amazing videos have been a huge help as I learn about NNs and Pytorch, thanks!!