How do you fit a Neural net with weight decay using the train function?

4 views (last 30 days)
I want to fit a Neural net for Regression and I'd like to use weight decay so I don't have to worry about using too many nodes. The neural nets I've fit so far have been with the train function and I don't see a weight decay option. I see one if I use trainingOptions with trainNetwork, but that's appears to be mostly for image classification and requires defining each of the layers of the network.
I'm hoping there is something like this available.
netOptions = fitnet(10, 'trainbr');
netOptions.trainParam.weightDecay = 0.1;

Answers (1)

Greg Heath
Greg Heath on 28 Jun 2018
Well, have you tried it ?
See
help trainbr
and
doc trainbr
Thank you for formally accepting my answer
Greg
  1 Comment
Michael Williams
Michael Williams on 28 Jun 2018
I had already looked through those files (and now again) and they don't mentioned weight decay at all. I did find some documentation about regularization and it might be what I'm looking for. https://www.mathworks.com/help/nnet/ug/improve-neural-network-generalization-and-avoid-overfitting.html#bss4gz0-38 Have you used it at all? Does the goal parameter in [networkName].trainParam.goal need to be fine tuned?

Sign in to comment.

Categories

Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!