This message was deleted.

Hello! Sure, transfer learning is possible in neuralforecast. You can check the tutorial here: https://nixtlaverse.nixtla.io/neuralforecast/examples/transfer_learning.html Of course, having some initial data will usually yield better predictions. I hope this helps!

Ok I'll check this out. I'm asking because I need to create forecast for newly introduced products (say soda or snack) and this new product has some related time series in the dataset (other soda or snack). So I thought maybe some of the neuralforecast algorithms like NHITS might be able to learn from similar products and thus create forecast for this new product even though the now product has no sales data.