Hi everyone, I'm trying to use Nixtla to recreate one of the proposed model solution in the literature. I'm having difficulties to understand a few key concepts. Can someone explain what is the difference between windows based model and recurrent base models?

At a very high level, they are both for univariate models. BaseWindows generates the entire forecast horizon in one shot, while recurrent will do it one step at a time. For example, MLP uses BaseWindows, but LSTM uses Recurrent. What model are you trying to implement? Maybe we can take it on our end and release it for everyone 🙂

@Marco Thank you very much for the offering and help! Architechture is called TCN-ECANet-GRU, but I would not want you to lose time with this, because this architecture is not a general one for time series prediction, but rather suited to my use case, to predict solar irradiance. I am trying to learn more how to code models from papers is the ultimate goal here 🙂