Universal transformer in streaming time series data

[Sergio Ryan]

Has anyone ever tried to use a universal transformer or transformer with a streaming time series data not related to NLP? For example in system log entries or or some sort of data collected from sensor periodic measurements? Am I in the right way if I want to use universal transformer in T2T for modeling time series data and predicting a set of sequences in the future or classifying some sort of time series data other than natural languages?

[Aleksas Pielikis]

I was thnging to try TransformerEncoder to predict a malfunction in electric system from periodic measurements within a specific time period which is a binary classification problem in a similar fasion an IMDB sentiment classifier. But haven't tried it yet.

[Sergio Ryan]

Yes my problem resembles yours. I have a set of timeseries data that I need to classify in a multi-class classification fashion. I am trying to use the Timeseries Problem subclass now with a TransformerEncoder or the universal one. Haven't got it right tho.