Unsupervised clustering of univariate time series

I’m new to pytorch.

I have a 23-year time series of remotely sensed vegetation index data (as a data file, not images). It is a univariate dataset - 1 variable, 23 time steps - in n observations (rows) and 23 columns. My hope was to use an RNN as an autoencoder and use the bottleneck as input into a clustering routine. As I understand it though, my data do not fit the expected format for an RNN since I only have one variable.

I looked at TorchCoder (GitHub - hellojinwoo/TorchCoder: PyTorch based autoencoder for sequential data) and Sequitur (GitHub - shobrook/sequitur: Library of autoencoders for sequential data). Does pytorch itself have a way to format my data so I can treat time as 23 variables and use GRU/LSTM in an autocoder format?


you can use pytorch standard RNN module to acomplish that.
You have an easy example at the very end.