How can I optimize layer and neuron numbers and choose the best activation functions for my neural network?

I'm working with feedforwardnet to solve a fitting problem and I was wondering how to find the best combination of

• layer and neuron numbers
• activation functions
• training/validation/test data set division

This should be a hyperparameters optimization problem, but I don't find anything suitable for neural networks such as those obtained with feedforwardnet

Answer 1

[Matt J]

Perhaps with the Experiment Manager?

https://www.mathworks.com/help/deeplearning/ref/experimentmanager-app.html?s_tid=srchtitle_experiment_2

[Giuseppe Maria Piuma]

I'll be trying it soon, thank you!