Best Validation check number for MATLAB neural network

5 visualizaciones (últimos 30 días)
Jack
Jack el 4 de Sept. de 2014
Comentada: Jack el 5 de Sept. de 2014
I'm using 10-fold cross validation and patternent function for a binary classification problem in MATLAB. When I see neural network result window, in all trainings of neural network ( 80% training , 10% validation and 10% test with sample size 200~600 ) Early stopping is stopping my training process in iteration between 20~40. As you know the default value is maximum 6. What should i do about this problem? Should i increase maximum number of early stopping iteration checks?
Thanks.

Respuesta aceptada

Greg Heath
Greg Heath el 5 de Sept. de 2014
Editada: Greg Heath el 5 de Sept. de 2014
That is not necessarily a problem.
What error rates are you getting as you vary the number, H, of hidden nodes and sets of random initial weights?
I typically look at Ntrials = 10 different initial weight initializations for each candidate value of Hmin:dH:Hmax (numH~10).
Search in NEWSGROUP and ANSWERS
greg patternet Ntrials
Hope this helps.
Thank you for formally accepting my answer
Greg

Más respuestas (0)

Community Treasure Hunt

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

Start Hunting!

Translated by