Hi Pourya, happy to know you liked the article!

As for the classification task, I'm not quite sure. At first I don't think so, I'm not sure whether there could be some strategy for using causality from impact as a way to classify time series.

What might be helpful though is using the library tensorflow_probability.sts which is essentially the core of tfcausalimpact.

You might find some of its implementation quite useful and helpful for describing time series.

Hope it helps :)