Neural Chase v1.0

Dusan Kostic — Sun, 11 Oct 2009 19:11:19 -0000

Yes, Neural Chase has finally come to an end of its development cycle. Chase between cats and mouses has been checked and extensive testing of application behavior has been concluded. My master thesis is finally developed and within a few days I will get my master degree.
This version has lots of changes and has included lots of configurations to test cats and mouse behavior, so feel free to test this application any time!

Neural Chase v0.7 Released

Dusan Kostic — Mon, 31 Aug 2009 19:58:06 -0000

After month of pause, I made few adjustments to new version. This one has newer Selection methods which can be chosen from Configurator. New Selection operators include Stochastic-Universal Sampling and Ranking Selection and with old Roulette Wheel Selection make 3 selection choices.

Neural Chase v0.6 released

Dusan Kostic — Fri, 03 Jul 2009 13:43:45 -0000

New release of Neural Chase includes the most usable additions. There were major fixes in genetic algorithm, and massive optimizations and restructures to make this project very attractive. Newer version is worked to work good with mouses. There are still glitches with cats, but they will be resolved in future versions.
Check this version, bronto!

Configurator added

Dusan Kostic — Tue, 23 Jun 2009 10:27:22 -0000

Newest release now include configuration maker, so you can easily create your own set of parameters for the application. This configuration system needs to be tested a bit more.
Next thing to do is to improve algorithms since they might have flaw which need to be fixed.

Minor bug fixed xD

Dusan Kostic — Mon, 15 Jun 2009 14:48:51 -0000

I forgot to uncomment the line to display items in listview so I had to quick fix that xD.
Check 0.4b version

Newer version released with some good GUI

Dusan Kostic — Mon, 15 Jun 2009 14:25:50 -0000

This release now posses lots of interface improvements on main application window. Much better statistics is presented. New TimeSeriesPloter component graphically displays fitness and generation data. Specific mouse and cat can now be selected and focused on application ground. Main menu is added but it has limited functionality.
Configuration maker is next to code, and it will allow end user to experiment with project parameters in run-time without needed to recompile it every time.

Some good example finally

Dusan Kostic — Wed, 10 Jun 2009 21:56:34 -0000

Lots of stuff where changed. Most of the glitches fixed and I decided to put cats as optionally in project. Now it is more concerned on mouses chasing cheese.
Future versions will surely have user interface to tune ANN and GA at your disposal.
Also future versions will be having more statistical and graphical improvements.

Fixed problems!!

Dusan Kostic — Sun, 07 Jun 2009 13:13:12 -0000

Phew... Simple logical bugs made me a lot of work, but I managed to pull this off.
Lot of "uneasy" bugs were taken care off. First I made a mistake about what sensory data should mouses take. They never knew their position on playground, so it was impossible for them to find any cheese. Next I was fixing fitness function to give advantage lot more advantage to mouses with most eaten cheese.
One of the greatest problems were in selection operator for GA. Chromo Roulette technique seemed to statistically somehow support chromosomes with less fitness. That problem was very hard to locate because it required constant observation of mouse progresses over generations. Resolution was to simple make chromosomes sorted in ascending order. Descending order was logically the same as ascending, but statistically ascending order are better.
Last problem was in with normalisation of Neural Network inputs, and tweaking neuron functions with better parameters. Now mouses finally steer correctly and inteligentlly, but more data tweaking is required.
I will upload soon newest version to SVN as it has a lot of reprogramming.

Testing of the chase

Dusan Kostic — Thu, 04 Jun 2009 17:51:05 -0000

These days, I implemented finally a lot of really working stuff in the project. I managed to develop sensors for objects on ground and their normalization to be compatible with Neural Networks. Later ANN were finally putted into action, alongside with genetic algorithms. They fitted well together, but...
Unfortunatelly, cats and mouses didn't get smarter. I rechecked and tested ANN and they seem to work fine. GA were tested and they also seem to work fine. Problem were at the sensors which cats and mouses gets from the outside world. I putted myself into their position and found that with those sensors I wouldn't be able to survive also.
Now I assumed easier task: to see will mouses be smarter in catching cheeses alone, without distraction of cats. Cats are disabled for time being, and as soon as I make mouses work together, cats should be few minutes job only.
Latest debug version can be again found on SourceForge SVN.

Sensory Data added

Dusan Kostic — Mon, 01 Jun 2009 22:16:00 -0000

Before I can start implementing ANN and GA frameworks, some sort of sensory data must be added. Those data will provide "natural senses" for cats and mouses. I added data placeholders for sensory data and their normalization to vector for input in ANN. Next is actually getting information about cats and mouse senses.
Of course my current progress is available on SVN repository.

Recent posts to news