I’ve created and trained my network successfully, but I have always assumed I could use DumpWeights to store the result of the training and use LoadWeights to bring it back when working on a different tree. My problem is it also appears I need the original data to construct a new MLP to evaluate other data. I tried using the default constructor and also tried an empty tree, but the only way evaluate doesn’t return nan is if I reload the original tree in the constructor. I was under the impression that it was the weights that were important for evalutation and am not quite sure why it needs the original data when it is being supplied variables for each event in the params array. Is there anyway I can use loadweights and not have to resort to export and loose a lot of flexibility from my code?
Thanks, I hope this makes sense.