Re: neural networks output problem
- From: Greg Heath <heath@xxxxxxxxxxxxxxxx>
- Date: Thu, 8 Jan 2009 07:58:38 -0800 (PST)
On Jan 6, 6:54 am, "Juraj Petrovic" <juraj.petro...@xxxxxx> wrote:
Hello!
I have a problem usingneuralnetworks in Matlab. My patterns are cells with 180 rows where each element is from {0,1}. The output of the network should also be either 1 or 0.
After initialising a ffneuralnetwork with:
net = newff(train_patterns, train_values, [250 100 1], {'tansig' 'tansig' 'tansig'},'traingdm');
(syntax of this command varies trough different versions of matlab, at least the first part, but I guess that's not the problem)
Your syntax is incorrect for either version.
help newff
and training the net with:
net.trainParam.show = 5;
net.trainParam.epochs = 200;
the net always returns the same value from [0,1] no matter what are the inputs. Changing the number of epoches doesn't cause a change to this (I still get only one, always the same value as the output, no matter the inputs) nor does a change of the layers or numbers of neurons in them.
Any ideas what am I doing wrong?
1. See above
2. You assumed we are mind readers:
a. You didn't explain the problem you are trying to solve.
b. You didn't specify what version of the Toolbox you are using.
c. You didn't clearly specify the dimensionality of the inputs and
outputs
d. You didn't clearly specify the number of observations
e. You didn't include the training and output commands
3. Convert to matrix inputs
4. You only need one hidden layer
6. Your output activation function should either be the default
(purelim) or logsig.
7. Use the default training function (trainlm)
See some examples: Search the Google groups archives
with
greg-heath newff train sim
Hope this helps,
Greg
.
- References:
- neural networks output problem
- From: Juraj Petrovic
- neural networks output problem
- Prev by Date: Re: Disable Excel Macro
- Next by Date: Re: Get the size of a docked figure?
- Previous by thread: neural networks output problem
- Next by thread: Re: unzip/gunzip into memory
- Index(es):
Relevant Pages
|