Multivariate Input

[zwebb2]

Does this implementation support multivariate input? For example, I would like to test a multivariate time series dataset from the UCR database (FingerMovements). This data has a size of (316, 50, 28) which is (batch_size, sequence_length, features).

Does the pytorch implementation support this currently? And if not, what would be needed to add this? I'm thinking a conv2d layer replacing the conv1d is a good place to start, but the kernel size would likely need to match the number of features in one direction - i.e. using all the features and only sliding the conv kernel in the time direction. Let me know if you have any thoughts. Thanks