Question about neural network weights/biases initialization

2 visualizaciones (últimos 30 días)
Natalia Genaro
Natalia Genaro el 21 de Abr. de 2011
Editada: Greg Heath el 12 de Jun. de 2014
Hi,
I'm creating a feedforward backpropagation network this way:
net = fitnet(7);
Then I set some parameters:
net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};
net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'};
net.divideFcn = 'random'; % Divide data randomly
net.divideMode = 'sample'; % Divide up every sample
net.divideParam.trainRatio = 185/289;
net.divideParam.valRatio = 35/289;
net.divideParam.testRatio = 69/289;
net.initFcn = 'initlay';
net.layers{1}.initFcn = 'initnw';
net.layers{2}.initFcn = 'initnw';
net.inputWeights{1,1}.initFcn = 'rands';
net.layerWeights{1,1}.initFcn = 'rands';
net.biases{1,1}.initFcn = 'rands';
net.trainFcn = 'trainlm'; % Levenberg-Marquardt
net.performFcn = 'mse'; % Mean squared error
After that, I initialite the network as follows:
net=init(net);
And when I want to access the weights and biases matrix:
net.IW{1,1}
net.LW{1,1}
net.b{1,1}
I just get the following answer:
Empty matrix: 7-by-0
I don't want to start training the net with all the weights set to zero. They should be random values, shouldn't they? How can I get them set to random values? Didn't I already do it when setting
net.inputWeights{1,1}.initFcn = 'rands'?
Thank you in advance

Iniciar sesión para responder a esta pregunta.

Respuesta aceptada

Greg Heath
Greg Heath el 9 de Jun. de 2014
Editada: Greg Heath el 12 de Jun. de 2014
net = configure(net,inputs,targets);
help configure
doc configure
Hope this helps.
Greg

Más respuestas (0)

Categorías

Más información sobre Define Shallow Neural Network Architectures en Help Center y File Exchange.

Etiquetas

Community Treasure Hunt

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

Start Hunting!

Translated by