Hello everyone! I have a large dataset of time series and I want to create embeddings for these time series to use in more classical models, as I have a small amount of data for regression. What are the best ways to compress large time series data (approximately batch_size x 1000 x 12) down to 10-16 features? I have tried using the hidden state of an LSTM and got decent results, but I would like to improve them. Thank you all!