Using Encog in Java

I'm running a backprop net (41x10x2) with 12,000 data points. I seem to quickly max out at 8 GB or more. Is this normal? Is there a way to reduce the memory footprint without changing the size of the problem?

A bug report -- this is strange in that I cannot be the first trying to create a PNN for classification on Windows VISTA...

Steps to produce the error:
1. File->New file
2. Choose Machine Learnign Method, type some name, click OK
3. In next dialog, choose PNN/GRNN, click OK
4. In the next dialog, put/choose something, anything, click OK, and you'll get the error

BTW, I tried to download the workbench source but the latest I can find is 2.5.3 in Google Code. Tried git, but the tarball has a few dozen java files in the root dir. Is there a better way to get the workbench source code?

Hi,

I am pretty new to NN, but I have a problem which I intend to solve using neural networks:

Basicly, I have an example linked list that contains an objects with timestamp and a hand position in 3d(x,y,z), I need to compare it to another linked list that also contains and object with timestamp and hand position(x,y,z). The original and new one might vary in size. I could easily record multiple example gestures of the same type in order to improve accuracy.

Hi all,

Could you please help me with some basic questions?

- How can I easily normalize a given MLDataSet (0-1 normalization)?
- Is it possible to normalize each field independently?
- A BasicLayer construcion with a null ActivationFunction assumes an ActivationLinear function?
- How default getError (a % error I assume) is calculated? What represent a 100% error?

Hi,
I am having trouble figuring out how to train Elman's network (like the one in recurent.elman.ElmanXOR example) with multiple sequences.

Sticking to the XOR example, in the util.TemporalXOR there is somethign like that:
public static final double[] SEQUENCE = { 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0 };
public static final double[] SEQUENCE2 = { 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0 };

I have a problem when build an elman recurrent neural network using encog 3.0 library. I got NaN error after few iteration.

This is my code in MATLAB :

net=newlrn (minmax(usdjpytrain),[4,1], {'logsig', 'purelin'}, 'trainscg', 'learngdm', 'mse');
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-5;

I wanna to create recurrent neural network like above in java. by default, a learning rate value in MATLAB is 0.01 and momentum is 0.9. but I got NaN error if I use learning rate and momentum like that. this is my code in java :

public class MseTrain{

Hello,

Unfortunately I could not find the documentation for Encog's Genetic Algorithms. The forum, wiki and also your commercial book (which is pretty good by the way) does not contain detailed info about GA.

So here are my questions:

Question 1:
In SolveTSP.java:
genetic.setMutationPercent(MUTATION_PERCENT);
genetic.setPercentToMate(PERCENT_TO_MATE);
genetic.setMatingPopulation(MATING_POPULATION_PERCENT);
genetic.setCrossover(new SpliceNoRepeat(CITIES/3));
genetic.setMutate(new MutateShuffle());
What exactly are the listed settings for and how do they work?

Hi, I'm new to NN and GA stuff. Encog is looking great so far. I just want to ask if GA is also CUDA accelerated?
As far as I've seen, GA is transformed into some sort of internal NN while iterating.
To be more specific / for example: Is the provided TSP example capable of taking advantage of CUDA?

Kind regards
Chris

encog 3.0.1 source
I believe the following line in org.encog.ml.genetic.mutate.MutatePerturb is incorrect:


value += (perturbAmount - (Math.random() * perturbAmount * 2));


The correct code should be:

value += value * (perturbAmount - (Math.random() * perturbAmount * 2));


Otherwise, very small values are overwhelmed by the perturbation.