I am training a neural network model in R but the solutions are not converging during training. so as a result i am not getting the desired output during prediction (r2 coming below 1). Can anyone tell me how can I get a better training model? and if anyone can provide me a modified code of neural net in R?
It would be useful to understand first, whether training is expected to converge at all-not all associations can be represented by any network, which means that it is expected, in that case, that the algorithm will not converge. And even if it does converge, for a certain number of examples; it's not true, either, that a given network can represent an arbitrary number of examples, drawn from a rule; it has a finite capacity.
These notions must be understood first; it's not a question of modifying the code trivially.
These references might be useful: http://www.lptms.u-psud.fr/membres/mezard/Pdf/89_MN_JPA.pdf
Article Stability-capacity diagram of a neural network with Ising bonds
Did you normalize the data before feeding into the neural net for training or using the raw data as is (in form of integers/decimal numbers)?
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training