OMG Could you please stop with your stupid predictions, I think the forum has seen enough of those.However from what i see I predict that in 2 months only NTAI and AAI will be capable of surviving, all other AI's will be crushed beneath the weight of the two.
2 preview pics of AAI
Moderators: hoijui, Moderators
Ok, at least your reaction is nice :)
But you two (edit: three) are the only ones who actually work actively on an AI right now, so the prediction was really too obvious...
And no Im not jealous, I'm just taking on to many things to do
But you two (edit: three) are the only ones who actually work actively on an AI right now, so the prediction was really too obvious...
And no Im not jealous, I'm just taking on to many things to do
Last edited by jcnossen on 08 Nov 2005, 21:35, edited 1 time in total.
Well actually, at the point that I stopped working on JCAI it had better recovery from builders getting stuck, and it builded defense, so it won on brazillian battefield. I'm not sure how this is now though, since you have been working on NTAI after that point.But everything I predicted in the JCAI thread has come true ^_^
probably i should consider it beeing a complement that aai will crush all other ais - but i just showed you nothing more than two pictures. so keep cool and avoid predictions that are not supported by facts
besides zaphod surely is not jealous and he helped me a lot, without his advice it would have taken me much more time to get so far
besides zaphod surely is not jealous and he helped me a lot, without his advice it would have taken me much more time to get so far
*finishes twiddling the controls on the triaxx throne and gets back to the situation at hand*
ahem, I would like to know which version of NTAI that was that JCAI beat? And I am aware that since I first tested NTAI against JCAI in the day or two after i released it I have yet to see a JCAI vs NTAI game come out with JCAI ontop. So yes if you where idnependantly working on a version fo JCAI and battled it against a publicly released NTAI, then that would be unfair. Otherwise Each NTAI has been able to destroy the equivilant version nubmer of JCAI. Infact untill 0.25 that was the main requirement before I would release it other than any bugs.
And I say this about submarine not absed on future facts I ahve yet to recieve. The very words he's written is enough to show that whatever AAI is it wont be a copy of what came before it, and it wont be a botched attempt either.
Eitherway zaphod tutelage wont save his apprentice from the onslaught that lies round the corner MUHAHAHAHAHAHAHAHAHHAHAHAHAHA
ahem....
ahem, I would like to know which version of NTAI that was that JCAI beat? And I am aware that since I first tested NTAI against JCAI in the day or two after i released it I have yet to see a JCAI vs NTAI game come out with JCAI ontop. So yes if you where idnependantly working on a version fo JCAI and battled it against a publicly released NTAI, then that would be unfair. Otherwise Each NTAI has been able to destroy the equivilant version nubmer of JCAI. Infact untill 0.25 that was the main requirement before I would release it other than any bugs.
And I say this about submarine not absed on future facts I ahve yet to recieve. The very words he's written is enough to show that whatever AAI is it wont be a copy of what came before it, and it wont be a botched attempt either.
Eitherway zaphod tutelage wont save his apprentice from the onslaught that lies round the corner MUHAHAHAHAHAHAHAHAHHAHAHAHAHA
ahem....
Nice work with what you are doing. Add the info to the wiki, so everyone can see what its currentyl doing.
AI Page: http://taspring.clan-sy.com/wiki/AI
Your AI should be on the http://taspring.clan-sy.com/wiki/AI:AAI page. I would add the info myself, but I dont really know how your AI works yet.
Use this page to learn how to edit the wiki if you dont already know:
http://taspring.clan-sy.com/wiki/Help:Editing
And set your page out similar to the [http://taspring.clan-sy.com/wiki/AI:NTAI]other[/url] [http://taspring.clan-sy.com/wiki/AI:ZcAIn]AI[/url] pages.
Good luck with it!
AI Page: http://taspring.clan-sy.com/wiki/AI
Your AI should be on the http://taspring.clan-sy.com/wiki/AI:AAI page. I would add the info myself, but I dont really know how your AI works yet.
Use this page to learn how to edit the wiki if you dont already know:
http://taspring.clan-sy.com/wiki/Help:Editing
And set your page out similar to the [http://taspring.clan-sy.com/wiki/AI:NTAI]other[/url] [http://taspring.clan-sy.com/wiki/AI:ZcAIn]AI[/url] pages.
Good luck with it!
it is stable right now if you use it on maps with "sharp" metal spots (not on small divide), because cain's metal spot finder class had some problems with it (don't know if he released an updated version, but tell me if so)
apart from capturing enemy units i had no crashes, but i didnt test it on a huge number of maps...
i'll release it as soon as possible..
apart from capturing enemy units i had no crashes, but i didnt test it on a huge number of maps...
i'll release it as soon as possible..
If it helps, cains class finds every spot with great accuracy. The problem comes when it tries to deal with thigns such as extractor radii and moving positions so that a moho could be buil next to the mex or if the mex patch is unbuildable terrain wise. That is where the accuracy is lost, and where all the fluke positions come from.
But the core code that fidns the spots in the first place is perfect be it spring style OTA style or mixed type metal maps.
But the core code that fidns the spots in the first place is perfect be it spring style OTA style or mixed type metal maps.
-
GrOuNd_ZeRo
- Posts: 1370
- Joined: 30 Apr 2005, 01:10
I recently posted (without any code) about learning AI as well, but used some mathematical terms like categorification and bayes rule in mine. I was thinking on ranking as well, and I think it would be important to keep the following information, rather than just unit kills:
1) Damage done per weapon type
2) Weapon vs armor type (some units are weak vs lasers, etc)
3) Scouting information about which units are built by the enemy
4) Range information, accurately stated, the range between target and unit when damage was done. A less accurate method would be reading weapon ranges but the margin of error may be very small. I notice some units don't fire weapons despite being in range, and as the discrepancy scales, so does the margin of error.
5) "effective damage per second" which is based on damage per tick while units were in range.
6) Evaluate current base defenses vs enemy units based on damage comparisons (If the enemy has 30k hp of planes, how many turrets per unit of distance are needed?)
7) create a massive system of interrelated weighted values, which will increase the likelihood of building strong anti-air near the most valuable buildings (fusion, etc)
Integrating base strategies like the AA anti artillery deflection or LLT surrounded by dragons teeth. This would increase defense, knowing when to build such structures is probably dependent on 6 and 7.
and then of course, general building tips. Of course, 6 and 7 are probably difficult ways to accomplish some things we could do with hacks and tweaks, but if we don't want to hard code these features I think that is the minimum required for an AI to learn this information. Of course, auto learning is ideal, because the AI could be used for hundreds of Mods, and be an effective player in each. Option 6 can vary widely in effectiveness and math, the stored weighted value per unit based on cartesian distance (x,y grid) would be one simple method, but wouldn't create a "density map" of DPS until range and field of fire is also included. Option 7 is obviously a bayesian system (see: http://yudkowsky.net/bayes/bayes.html )
A way to gather statistics faster would be allow a method of submitting unit statistics. In my other (very recent) post, I made a general list of what I think a complete AI should have.
As for AI used in games, many do. But there are probably more unique concepts of what "Artificial Intelligence" is and what qualifies as an AI than are used in games. I play around with (and hope to create soon) Genetic Algorithms and Neural Networks as well. (Self organized map NN) Some claim these are AI, I don't think they are. Most game AI is reactive scripting based on simple hardcoded in game values. The real intelligence is the programmer, not the computer unit behavior.
1) Damage done per weapon type
2) Weapon vs armor type (some units are weak vs lasers, etc)
3) Scouting information about which units are built by the enemy
4) Range information, accurately stated, the range between target and unit when damage was done. A less accurate method would be reading weapon ranges but the margin of error may be very small. I notice some units don't fire weapons despite being in range, and as the discrepancy scales, so does the margin of error.
5) "effective damage per second" which is based on damage per tick while units were in range.
6) Evaluate current base defenses vs enemy units based on damage comparisons (If the enemy has 30k hp of planes, how many turrets per unit of distance are needed?)
7) create a massive system of interrelated weighted values, which will increase the likelihood of building strong anti-air near the most valuable buildings (fusion, etc)
Integrating base strategies like the AA anti artillery deflection or LLT surrounded by dragons teeth. This would increase defense, knowing when to build such structures is probably dependent on 6 and 7.and then of course, general building tips. Of course, 6 and 7 are probably difficult ways to accomplish some things we could do with hacks and tweaks, but if we don't want to hard code these features I think that is the minimum required for an AI to learn this information. Of course, auto learning is ideal, because the AI could be used for hundreds of Mods, and be an effective player in each. Option 6 can vary widely in effectiveness and math, the stored weighted value per unit based on cartesian distance (x,y grid) would be one simple method, but wouldn't create a "density map" of DPS until range and field of fire is also included. Option 7 is obviously a bayesian system (see: http://yudkowsky.net/bayes/bayes.html )
A way to gather statistics faster would be allow a method of submitting unit statistics. In my other (very recent) post, I made a general list of what I think a complete AI should have.
As for AI used in games, many do. But there are probably more unique concepts of what "Artificial Intelligence" is and what qualifies as an AI than are used in games. I play around with (and hope to create soon) Genetic Algorithms and Neural Networks as well. (Self organized map NN) Some claim these are AI, I don't think they are. Most game AI is reactive scripting based on simple hardcoded in game values. The real intelligence is the programmer, not the computer unit behavior.
