Feedforward network using backpropagation in Encog

659 views Asked by At

I am using this classification example by jeff heaton:

https://github.com/encog/encog-java-examples/blob/master/src/main/java/org/encog/examples/guide/classification/IrisClassification.java

I am trying to find out what kind of learning algorithm and how many hidden layers will be used in this example. I couldn't figure this out.

1

There are 1 answers

0
JeffHeaton On

For a learning algorithm I would suggest using RPRop, it is one of the most general purpose algorithms provided by Encog. You can also try Levenberg Marquardt, but it does not scale as well with large data sets.

Hidden layers are essentially hyper-parameters, and finding the the right set could take some trial and error. I suggest a single hidden layer with (num_inputs + num_outputs) * 1.5. This is just a starting point, you can increase or decrease as needed.