My goal is to learn latent representations of a multivariate time series which would be used for supervised learning experiments later. I believe an LSTM autoencoder is a good w