how to get objective function in neural network
3 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
prem kumar
el 25 de Mzo. de 2015
Comentada: prem kumar
el 27 de Mzo. de 2015
my question:
i have five inputs A B C D E and two outputs X Y. [A B C D E] are in the form of numerical data that is [ 115 47 21 3 4; 115 47 21 3 5; 115 47 21 3 6; 115 47 21 4 3; .............n27]. this 27 experiments conducted based on orthogonal array form. the outputs are [X y] is [ 3.41 6.24 3.25 6.1 2.94 5.14 .........n27].
i have to form neural network for these data after successful training i have to predict the unknown data accurately. after training the neural network its possible to get weight and biases value. till now i complete my neural network training in matlab application tool box i.e.. nntool.
in this i separate 22 data for training 5 data for testing. i used 2 hidden layer and 6 hidden neuron. my network is based on back propagation neural network with sigmoid and pure linear transfer function. i set the training epoch as 1000 and max fail is 6.
whether its correct or not? but after training the regression curve are quite better. and also after training is achieved my MSEis 0.00092.then i predicted the remaining 5 data, and the out put of those data are quite match with that my experimental data. how to fix the mse 0.0000001?.
whether my work is correct?. if correct means using the weights and biases value how to form an objective function in this trained neural network. because optimize this objective function with genetic algorithm. this works seams to be a hybrid of neural network with genetic algorithm.
whether its possible? if means how i do ? how i get the objective function? my work is purely on mechanical domain.
plz help me ...
0 comentarios
Respuesta aceptada
Greg Heath
el 25 de Mzo. de 2015
Editada: Greg Heath
el 25 de Mzo. de 2015
I think you are making the problem harder than it is. For example, neither a second hidden layer nor a genetic solution is warranted. Genetic solutions are not supported in the NNToolbox and such a solution may take orders of magnitude longer than straightforward backpropagation via trainlm.
The documentation for regression is obtained via help fitnet doc fitnet
Also see my posts on FITNET regression in the NEWSGROUPand ANSWERS. Search on
greg fitnet
greg fitnet tutorial
The latest was posted in the NEWSGROUP an hour or so ago.
Hope this helps.
Thank you for formally accepting my solution
Greg
PS The objective function is the function that is minimized to obtain the weight solution. Is that what you really want?
4 comentarios
Más respuestas (0)
Ver también
Categorías
Más información sobre Function Approximation and Clustering en Help Center y File Exchange.
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!