I have a problem where I train a fully connected NN and once trained I gather new data and train again (with both old and new data) for a few times.
Do I need to reinitialize something in order the model to learn correctly? Optimizer or the model weights. I guess if I use
lr_scheduler I should reinitialize but I do not use it yet.
I use ADAM optimizer.
Thanks in advance!
(New to Pytorch)