Is it possible to train a pytorch LSTM with a sequence containing several features, but a single label?
If so, how does one format the data, and what size is the input for the LSTM?
time1 feature1, feature2, feature3, feature4
time2 feature1, feature2, feature3, feature4
time3 feature1, feature2, feature3, feature4, label
Assume the data is in a csv, each row is a time step and contains a single sequence of 4 features, and every 3 rows has a single label.
I’ve tried asking this question several different ways on this forum and never get a response. So either i’m asking the question poorly, or no one understands what i’m asking. I could really use some help. For reference, i’ve read (i’m fairly sure) all of the pytorch tutorials on the lstm topic. None of them seem to address what i’m trying to do, which is use several sequences with a single label, instead of labeling every sequence. That’s because in my case, we have a cause, and later, an effect, and they are displaced in time (a time series). The classification isn’t known until the end of the sequence, thus, several sequences per label. Importantly, every sequence has several features.
Is this possible? What’s the proper name for this kind of approach?
Any help would be very much appreciated. Thank you.