I am working on a time series project with hospital data in the ICU. The admissions have variable durations and the data is recorded at minute intervals (regular time steps). The data has already been imputed and interpolated, and the missing values have been handled.

I want to train an LSTM model with this data, but I am unsure how to preprocess my input. I want to create tensors, but how should I handle the varying lengths of admissions?

I was thinking of using separators between each admission, but what would be the best strategy in this case?

Thank you for your insights and experience.