How to code up Neural Networks ?
1 view (last 30 days)
Show older comments
I am new to Neural Networks. I have to code up whatever is written in below image.
Is there a need of nprtool to do that?
Accepted Answer
Greg Heath
on 12 Feb 2014
I suggest using a standard classifier with patternnet, one hidden layer, and target columns from eye(23).
Use as many defaults as possible.
Search using
greg patternnet
for examples.
Hope this helps.
Thank you for formally accepting my answer
Greg
More Answers (2)
Greg Heath
on 12 Feb 2014
I am confused by your question
"Is there a need of nprtool to do that?
Please clarify that statement.
Comment
The figure shows a 3-layer feed-forward multilayer perceptron neural network with three neuron layer outputs ai (i=1:3).
The input box indicates a layer of non-neuron fan-in-units.
There are two hidden neuron layers and an output neuron layer.
One hidden neuron layer is sufficient for a universal approximator.
One hidden layer FFMLPs are the "standard" configuration
The standard output for a 23-class classifier/pattern-recognizer is a 23 node output layer. The net is then trained with column vectors from the 23-dimensional unit matrix eye(23).
So, are you claiming that the extra hidden layer and single output learns faster and better than the standard classifier with one hidden layer and 23 outputs?
1 Comment
Greg Heath
on 15 Feb 2014
Once you have the input matrix of N I-dimensional column vectors and target matrix of N c-dimensional column unit vectors you can use the code from one of my patternnet postings.
Searchwith
greg patternnet
Hope this helps.
Thank you for formally accepting my answer
Greg
0 Comments
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!