我想知道在这种情况下如何将参数放在 nn.LSTM 中

Question

I want to say that I'm not familiar with English.

Hi, I'm recently building LSTM Layers using nn.LSTM .

What I want to do is predicting next 48 values using 48*7=336 sequential inputs.
And here is preprocessed dataframe will be used. It has 384 cols(=336+48) and 457068 rows(=929x492).
In short, there are 929 customers and each customers has 492 sequential data with a length of 384(=336+48)

customerID	col1	col2	...	col380
1	data_{1,1}	data_{1,2}	...	data_{1,380}
1	data_{2,1}	data_{2,2}	...	data_{2,380}
...	...	...	...	...
1	data_{492,1}	data_{492,2}	...	data_{492,380}
2	data_{493,1}	data_{493,2}	...	data_{493,380}
...	...	...	...	...
929	data_{4670687,1}	data_{4670687,2}	...	data_{4670687,380}
929	data_{4670688,1}	data_{4670688,2}	...	data_{4670688,380}

Here, 3 parameters for nn.LSTM: (input_size, hidden_size, num_layers)
Also Input tensor has 3 dimensions: (seq_len, batch, input_size)\

In this situation, please let me know how to enter the above parameters. I will use num_layers=2 .
Thank You.

Answer 1

Well, in the 3 parameters for the LSTM, input_size will be the same as input_size in your input tensor, which here is 336 .

hidden_size could be any number you want, it is basically the number of nodes or features you want to learn. The higher the number the more the model learns about your data at each training iteration. It mostly depends on how tough your task is for your model, though you would not want this to be unnecessarily large.

Since you have chosen num_layers=2 you can define the LSTM like this:

# Using hidden_size=256, which could be any number you want
nn.LSTM(input_size=336, hidden_size=256, num_layers=2)