I have one large file that has one measurement each second for 100,000 seconds. Each measurement captures 3 features.
I am going to create an LSTM that will look at 30 seconds in one go. i.e.
seq_len = 30, input_size = 3.
As I have a single file with continuous data measurements being taken for 100,000 seconds, after passing in the first set of data points in to this lstm of len 30, I am going to pass on the hidden state when examining time points 31 to 60.
If I wanted to batch this process, to look at two different 30 second windows at the same time, is there a DataLoader that I can use to ensure that between each seq_len, the data is correlated, but between different batches, the data is independent?
so my input size will be
(seq_len=30, batch=2, input_size=3)
two hidden states will be passed on to the next two batches, and the next two set of seq_lens are from time points that follow the first two sets of seq_lens.