Recent changes to support-requests

Bug Query: Monitor.java

Jonathan Love — Thu, 14 Apr 2005 16:00:26 -0000

Occasionally when I've used Joone, I've received this
error:

Exception in thread "Thread-5608"
java.lang.ArrayIndexOutOfBoundsException: 1265
at
org.joone.engine.Monitor.fireNetStopped(Monitor.java:192)
at
org.joone.engine.NetStoppedEventNotifier.run(NetStoppedEventNotifier.java:26)
at java.lang.Thread.run(Thread.java:595)

I have no idea what the cause is but I've discovered if
I change org/joone/engine/Monitor.java, line 191:

From:
for (int i = 0; i < size; ++i) {
To:
for (int i = 0; i < list.length; ++i) {

The error doesn't seem to occur. I was also originally
confused as to why use ++i, not i++?

I'm curious what the cause of the error is - I don't
have the greatest understanding of the Joone internals
so haven't been able to work out the cause.

Validation, execution bugs with large net

Paolo Marrone — Thu, 18 Nov 2004 09:39:12 -0000

I am able to reliably reproduce the following two bugs.

I am using Joone 1.1.1, the network is at
http://www.cs.utexas.edu/~surdules/ids-elman-train-10p.ser,
and the training data is at
http://www.cs.utexas.edu/~surdules/kddcup.data.10percent.joone.

Bug #1

1. Load the network in the GUI
2. Click Tools->To Do List
3. The network is valid, dismiss the dialog
4. Click File->Save
5. At this point, the GUI either hangs with 100% CPU
utilization, or sometimes it saves a 22MB .ser file

Bug #2

1. Load the network in the GUI
2. Click Tools->Control Panel
3. Click Run
4. At this point, the GUI hangs with 100% CPU
utilization. This is surprising especially considering
that I am only using 10 training examples and 2
training epochs!

Thank you for any help you can provide with these
problems,

Razvan.

Adding new InputSynapse in GUI Editor

Eric — Wed, 06 Oct 2004 20:24:12 -0000

I am trying to add a new InputSynapse to the GUI
editor.I see the layers.xml file seems to hold the GUI
specs, but editing the file in either the joone install
directory, or the one in the joone-editor.jar file didn't
display the new synapse.

Am I assuming incorrectly that the layers.xml file is used
to determine the icons to display/object to use in the
editor? Or was this file used to create a compiled java
class and I will need to recompile the GUI editor?

Thanks

Eric Foertsch

Using Excel Input layer with formulas-get zeros

Eric — Sat, 28 Aug 2004 18:13:51 -0000

Not sure if this is in the Joone domain or not. I have an
Excel spreadsheet with 2 worksheets. The first
worksheet has values. The values on the second
worksheet all are based on formulas going against the
first worksheet. When I use an Excel input layer in the
Joone GUI Editor, use the second worksheet as input
and then 'inspect' the values the editor shows all zeroes.

Is this a limitation of the Excel input layer or a bug?

Also want thank all the developers for providing a great
piece of software. I always wanted to learn about NN
and this software really helps.

Thanks

How do I stop the threads

Sam — Fri, 02 Jul 2004 16:51:07 -0000

Hi,

I am using DirectSynapses for my inputs and outputs.
When I try stopping the threads using the stop method
they are still running. Also, I am getting repeated
outputs. e.g. my 3rd output is the same as my 2nd. I
have tried using the code in the xor example:

Pattern stop = new Pattern(new double[2]);
stop.setCount(-1);
memInp.fwdPut(stop);
memOut.fwdGet();

but this gives repetition as well. This repetition is not
correct because if I swap the input patterns around I
still see the same outputs repeated.

I would be extremely grateful if someone could take a
look at my code and let me know how to fix this problem
I am seeing.

Thanks,

Sam

My code is below:

import org.joone.engine.*;
import org.joone.io.*;
//import org.joone.engine.learning.TeachingSynapse;
import org.joone.net.*;
import java.io.*;

public class TestOthello {

// The input array used for this example
private String snetFile
= "C:\\Sam\\assignments\\dissertation\\Joone\\othello.sn
et";
private NeuralNet othello;
private DirectSynapse memInp;
private MemoryOutputSynapse memOut;
Object networkLock=new Object();

public TestOthello() {
}

// Method that loads the neural net
public void loadNet() {
// We load the serialized XOR neural net
othello = restoreNeuralNet(snetFile);
if (othello != null) {
/* We get the first layer of the net (the input layer),
then remove all the input synapses attached to it
and attach a DirectSynapse */
Layer input = othello.getInputLayer();
input.removeAllInputs();
memInp = new DirectSynapse();
input.addInputSynapse(memInp);
/* We get the last layer of the net (the output layer),
then remove all the output synapses attached to it
and attach a MemoryOutputSynapse */
Layer output = othello.getOutputLayer();
// Remove all the output synapses attached to it...
output.removeAllOutputs();
//...and attach a MemoryOutputSynapse
memOut = new MemoryOutputSynapse();
output.addOutputSynapse(memOut);
System.out.println("Network loaded");
}
}

public void start() {
othello.getMonitor().setLearning(false);
othello.start();
}

public void stop() {
Pattern stop = new Pattern(new double[64]);
stop.setCount(-1);
memInp.fwdPut(stop);
memOut.fwdGet();
}

// Method to get output of neural net
public double getOutput(double [] in) {
// Now we interrogate the net
//othello.getMonitor().setTotCicles(1);
//othello.getMonitor().setTrainingPatterns(4);

othello.getMonitor().setLearning(false);
othello.start();

//othello.getMonitor().Go();
// Prepare the next input pattern
Pattern iPattern = new Pattern(in);
iPattern.setCount(1);
// Interrogate the net
memInp.fwdPut(iPattern);
// Read the output pattern and print out it
//double [] pattern = memOut.getNextPattern();
Pattern pattern = memOut.fwdGet();
double out = pattern.getArray()[0];
Pattern stop = new Pattern(new double[64]);
stop.setCount(-1);
memInp.fwdPut(stop);
memOut.fwdGet();
/*iPattern.setCount(-1);
memInp.fwdPut(iPattern);
memOut.fwdGet();*/
//othello.stop();
return out;
}

// Method to read in a serialized neural net object
public NeuralNet restoreNeuralNet(String fileName) {
NeuralNet nnet = null;
try {
FileInputStream stream = new FileInputStream(fileName);
ObjectInputStream inp = new ObjectInputStream
(stream);
nnet = (NeuralNet)inp.readObject();
}
catch (Exception excp) {
excp.printStackTrace();
}
return nnet;
}

public static void main(String[] args) {
double out;
TestOthello othello = new TestOthello();

othello.loadNet();
//othello.start();
// pattern 1
double [] in1 =
{0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.
5,0.5,0.5,0.5,0.5,0.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,1.0,0.0,0
.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,
0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
,
0.5,0.5,0.5,0.5,0.5};
out = othello.getOutput(in1);
System.out.println("Output Pattern #"+(1)+" = "+out);
//out = othello.getOutput(in1);
// pattern 2
double [] in2 =
{0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.
5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.0,0
.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,
0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
,
0.5,0.5,0.5,0.5,0.5};
out = othello.getOutput(in2);
System.out.println("Output Pattern #"+(2)+" = "+out);
//out = othello.getOutput(in2);
// pattern 3
double [] in3 =
{0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.
5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,1.0,0.0,0
.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.0,0.5,0.5,0.5,0.5,0.5,0.5,
0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
,
0.5,0.5,0.5,0.5,0.5};
out = othello.getOutput(in3);
System.out.println("Output Pattern #"+(3)+" = "+out);
//out = othello.getOutput(in3);
// pattern 4
double [] in4 =
{0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.
5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,1.0,0.0,0
.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,
0.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
,
0.5,0.5,0.5,0.5,0.5};
out = othello.getOutput(in4);
System.out.println("Output Pattern #"+(4)+" = "+out);
//out = othello.getOutput(in4);
// pattern 5
double [] in5 =
{0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.
5,0.5,0.5,0.5,0.5,0.5,1.0,0.5,0.5,0.5,0.5,0.5,0.0,0.0,0.0,0
.5,0.5,0.5,0.5,0.5,0.5,0.0,1.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,
0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
,
0.5,0.5,0.5,0.5,0.5};
out = othello.getOutput(in5);
System.out.println("Output Pattern #"+(5)+" = "+out);
//out = othello.getOutput(in5);
}
}

Bug in batchmode?

Boris Jansen — Wed, 30 Jun 2004 09:36:07 -0000

While I was looking at some code, I think I have spotted
a small error in org.joone.engine.BatchLearner.

In the method requestBiasUpdate, in the for-loop, the
object myBias is updated myBias.value[x][0] +=
myLearningRate * ... and it even has the
comment " //local update of myBias, not yet given back
to the Layer ", where a little further, after the loop the
bias is set in the layer; myLearnableLayer.setBias
(myBias); with comment " // batch fullfilled, so update
the bias now and start over ".
The problem is in initiateNewBatch() myBias is set as
follows: myBias = myLearnableLayer.getBias(); I think
myBias should be set to a clone/copy of
myLearnableLayer.getBias(); else one works one the
bias of the layer itself directly in requestBiasUpdate. And
so it isn't a local update as is written in the comment...

Date format in Yahoo Finance

Anonymous — Sun, 16 May 2004 21:38:24 -0000

Hello -

I am trying out the provided Turning Points/Yahoo
Finance Input sample. My problem is that I can't get
dates in US format to work.

I looked at the source code and I thought dates such as
May 5, 2004 should work. But I keep getting a validation
error msg on the 'To Do' list of Error -
YahooFinanceInputSynapse - InputLayer 1 - Format for
End Data is invalid". I have tried various other
combinations (mm/dd/yyyy, dd/mm/yyyy, mm/dd/yy )
with equally bad results.

Any ideas?

Thanks in advance

Eric Foertsch

FIR example

Jerry W. Johnson — Mon, 14 Apr 2003 21:24:31 -0000

Hello,

I'm trying to duplicate the HENON MAP example on
http://www.cs.hmc.edu/courses/1999/fall/cs152/firnet/de
mo1.html.

I built the following network using the recommended
learning parameters and the henon map data:

fileInp -> layer,1row -> [-1,4taps] -> layer,7rows ->[-
1,3taps] -> layer,1row <- teacher <- fileInp

This network trains to an MSE of .009 after about 150
epochs. My question is: After training, how do I
implement Wan's method of using the output of the last
layer as input for the next step? Just connecting the
output layer both to the input layer and an output file,
and clicking run on the control panel doesn't seem to do
anything. The computer bogs down with CPU usage, but
the output file never gets any data.

Do you have an example you can send me?

Thank you,
Jerry W. Johnson

Kohonen layer

Hein Beukes — Tue, 25 Mar 2003 10:03:31 -0000

My input file: BWNeural.txt

Band1 Band2 Band3 Training
179 99 54 1
182 76 49 1
34 88 101 0
190 171 150 0
171 84 50 1
.
.
.

A thousand entries (data of a sample) of which only 130
got a entry of 1 under the column 'Training'. The values
of '1' represent areas to be extracted from a Landsat TM
image with 3 colour bands. Obviously it is easily done
with image processors but I need to get acquainted with
Joone.

Is it possible to put me (giving all is normal) on tract?
The intention is to assign values of '1' (where applicable)
to the rest of the image (millions of entries). What about
the Kohonen layer for classification?

Regards
Hein

Error with XOR sample

William D Powell — Thu, 28 Nov 2002 07:43:12 -0000

I downloaded the latest version 0.98. I put the joone-
engine.jar on my classpath. I ran the following
command and got the resulting error messages. What
have I done wrong?
[root@lenz joone]# java
org.joone.samples.engine.xor.XOR xor.txt error.txt
Training...
java.lang.NullPointerException
at
org.joone.engine.learning.TeacherSynapse.forward
(TeacherSynapse.java: 70)
at org.joone.engine.Synapse.fwdPut
(Synapse.java:122)
at
org.joone.engine.learning.TeacherSynapse.fwdPut
(TeacherSynapse.java:1 01)
at
org.joone.engine.learning.TeachingSynapse.fwdPut
(TeachingSynapse.java :28)
at org.joone.engine.Layer.fireFwdPut
(Layer.java:166)
at org.joone.engine.Layer.run(Layer.java:312)
at java.lang.Thread.run(Thread.java:536)