I am trying to get a model to predict the concrete expansion in aggressive environments (output), by using 7 inputs, but I don't know wich activating and transfer functions to use.
Hi,
If you want to learn the principles of ANN there are lots of books, but if you want only learn how to develop a model in MATLAb you can use MATLAB's help. It is the best for learning how you can develop your own model. Also, you can use MATLAB neural networks toolbox.
To develop your own model instead of matlab toolbox write: help newff
You can change activation functions and learning algorithm, the number of hidden layers and neurons ,etc
I suggest: Neural Network Design, Martin T. Hagan, Howard B. Demuth, Mark H. Beale
There is a good book: "Neural Networks - A comprehensive foundation" - Simon Haykin. But you can also follow the Matlab's help, there are several examples that you can use.
Thank you, i already red the matlab's help, and at the moment i'm working on a special algorithm not the toolbox of matlab, this algorithm let me even draw the variation of the output with other inputs! But my ignorance is the impact of changing activation functions on the model fiability, and what are the domains of each one of them.
I also suggest, initially use existing transfer functions and algorithms available in matlab, use sigmoid type transfer functions in hidden layer (s) units and linear TF in output unit. It is better to normalize the data. 1. Neural Network Design, Martin T. Hagan, Howard B. Demuth, Mark H. Beale, 2. Practical Optimization by Gill, P. E.; Murray, W.; Wright, M. H.. First try `''trainbr'' if not getting smaller error then go for trainbfg or traincgb. Best of luck
i always use ''sigmoid'' for the HL and ''purelin'' for the OUTPUTL., trainbr it's also my choice and sometimes ''tranlm'' but never 'trainbfg'' or ''traincgb''. i'll try them and see the difference in results! Thank you.
you can use tanh() function as an activation function from input node to hidden nodes and an Identity function may be used as an activation function between hidden nodes to output node. Since tanh() function is more efficient than the sigmoid function.
I think that ANN Matlab toolbox is a very friendly and easy to use solution. The book from Simon Aykin -suggested by the colleague from Brasil, Sao Paulo (SP)- will add the necessary theoretical background
- Badges
- Science topic
- Similar topics
- Engineering
- Materials Engineering