This is an archive of a previous version of Sodarace.

Please visit http://sodarace.net for the latest version.

forum  |  »» sodaplay
 
»» forums  »» artificial intelligence forum

subject: AI help thread

11 replies on 1 page. most recent reply: 01-Aug-07 02:36 by RJRocket53

»» back to topic list  

This topic has 11 replies on 1 page
»» previous topic   »» next topic  


mmaarrkkuuss

»» models
»» homepage

AI help thread   posted: 13-Apr-07 22:01   »» 
If you need any help on AI you can ask here. This thread has almost the same rules as in the model maker help thread.

mmaarrkkuuss



John_F

»» models
»» homepage

Re: AI help thread   posted: 16-Apr-07 22:54   »» 
Very good idea for a thread mmaarrkkuuss, though this forum does seem very dead.

woffwoff2

»» models

Re: AI help thread   posted: 05-Jun-07 00:59   »» 
can anyone tell me where you can get started in AI? im sorta new

florianseidel

»» models

Re: AI help thread   posted: 27-Jul-07 18:49   »» 
Hello,

I could need some help. Have been working on Neural Nets during the last days and am completely losing my mind over it. I start doubting my skills. I implemented a 2 layer feed forward network with sigmoid threshold funciton which should be trained using stochastic backpropagation to lern a simple xor function.
Source in C++ is attached and should be easy to understand.
The problem is: it is not working, i.e. not converging to a usefull state even after millions of training iteration. Please, it must be something simple, anyone have a look at it! PLEASE, I am losing it!!! Have been looking for an error for days....
Thanks in adavance.

file: http://sodarace.net/upload/florianseidel/Backpropagation.cpp

RJRocket53

»» models
»» homepage

Re: AI help thread   posted: 27-Jul-07 21:25   »» 
I need help with neural nets and A* Path finding. I am programming a game were you you program a robot that plays tag, the programming will be neural net type

cactus2

»» models

Re: AI help thread   posted: 27-Jul-07 23:32   »» 
florianseidel, currently looking through the code, but I haven't found any errors.

RJRocket, I'm not sure if neural networking is the best way to make a robot play tag. There are plenty of simpler, more effective, ways.

RJRocket53

»» models
»» homepage

Re: AI help thread   posted: 30-Jul-07 04:24   »» 
I mean that in the program, you will program the robot, you don't control it. You will design the neural net to program the robot.

cactus2

»» models

Re: AI help thread   posted: 30-Jul-07 13:12   »» 
A neural net used to control a robot would not be very useful. A neural network's advantage over convential algorithms is in recognization of various things, such as handwriting, faces, or the like, and that requires hundreds of nodes, as well as many epochs of training. A regular, iterative script-based loop would be prefered. This would allow for more flexibility than a net.

RJRocket53

»» models
»» homepage

Re: AI help thread   posted: 30-Jul-07 16:02   »» 
but I don't know how to make a program that you you program in

RJRocket53

»» models
»» homepage

Re: AI help thread   posted: 30-Jul-07 21:31   »» 
Sorry, it's not really tag, there is health, and your player loses 10% of it when it gets tagged, also, when you get tagged you don't become it, there is a timer that will switch the person who's it every 10 seconds or so. there will be tile like things, but they don't have pictures. Your player can shoot and it stuns your enemy if it hits it. You can push the blocks if they aren't two deep or a player is on the other side. The blue is you. The green is the computer. Red is the health. It reapears after it is gotten.

cactus2

»» models

Re: AI help thread   posted: 30-Jul-07 21:37   »» 
What language are you programming in?

RJRocket53

»» models
»» homepage

Re: AI help thread   posted: 01-Aug-07 02:36   »» 
VB, sorry but the attach file for thepicture won't work



This topic has 11 replies on 1 page


»» previous topic   »» next topic  

»» back to topic list  »» top of the page  

PLEASE READ THE FORUM GUIDELINES AND ALWAYS PREVIEW TO CHECK MESSAGES BEFORE POSTING...
...Help keep our forums creative and constructive. Thank you.


»» forum home