how to change nnstart's params in deep learning toolbox ?

13 vues (au cours des 30 derniers jours)
jinwoo lee
jinwoo lee le 18 Fév 2021
Commenté : jinwoo lee le 22 Fév 2021
Hello, All
This is a question on matlab's nnstart.
Is there a way to change activation function( to tanh), num of epoch and wieght initialization settings in nnstart?

Réponse acceptée

Shashank Gupta
Shashank Gupta le 22 Fév 2021
Hi,
Its not possible to change activation function from nnstart gui, You actually need to use fitnet or patternet function to initialise the model and then change the activation function there. Also you can decide on weight and bias initialisation in the same functions. Regarding the number of epoch, this functions generally stop training when validation loss become saturated. But this function does not provide flexibility to change the default number of epoch. If you wish to change a lot of things and define a lot of custom function to tackle you problem, I would recommend you to use deep learning framework such as trainNetwork or dlnetwork, These are more advanced and more flexible in choosing hyper parameters.
I hope this helps you,
Cheers

Plus de réponses (0)

Community Treasure Hunt

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

Start Hunting!

Translated by