Actual feature :
- codesize 603 - momentum - bias training - any structure that can be described by an int array
the library can be found in the MiniNeural package here https://robocoderepository.com/BotDetail.jsp?id=1754
New version 1.2 can be found here : https://www.geocities.com/synnalagma/neuralib.html
New features :
- codesize 591 - contain comments to reduce codesize until 500 (by removing feature like momentum, bias or just changing some code) - contain new IO functions much smaller, thanks to Kawigi (I don't use them so...) - contain float version (shrink file size) same codesize - contain XOR exemple with file IO
I'm quite sure Albert will make a killer with this. --Synnalagma
Je, je. Nice to know that now I can build a NeuralMiniAspid?. -- Albert
Your ScruchiPu bot is just the one I'm trying to beat since I started with NeuralNetwork and still nothing --Synnalagma
With the I/O functions in the NeuralNetwork class, what would be wrong with just serializing the NeuralNetwork object? -- Kawigi
Nothing but in this case you must do an init methods in the NeuronLayer? class, you're right I'ill do it this way, thanks for the comment --Synnalagma
If there was a lot of other stuff you have to do, you can also just write in writeObject and readObject methods into those classes (instead of your save and load methods) and they would get called instead of automatically serializing the objects. Of course, I always think of that as being a sort of megabot thing (FloodHT does it, for instance, FloodMini doesn't). -- Kawigi
You can shrink a little more the codesize by removing some attribute's property (private, protected) and also by using clone() method (see comments in NeuronLayer?.java), I don't like doing this so I didn't. --Synnalagma