OTAI Thread
Moderators: hoijui, Moderators
Corner Attack Bug: Seen it on Ashap Plateau. It's caused when the Attack Team has no targets and uses the default (0,0,0). I'm fixing it by having the Attack Team reset it's counter so that it waits instead of charging off to the great north-west.
CPU Peak: I still don't know for sure why, but I suspect that it is related to searching for a building location, as it tends to crop up when they are not available. I'm considering doing this kind of work in a seperate thread so that it doesn't impact performance so much.
This also might be because there are too many metal spots, as on wide open combat, or that asteroid one (what's it called?).
Support for FF: I've downloaded the mod, but I haven't really played with it yet. I'll get some support for it, but I'm not that great a player, so I can't promise much.
Also, there is a bug when the first unit of a new Attack Team is a gunship, the AI crashes because air units don't have move information. I'm changing this to count the move type as hover for now.
For 1.08, I'm also relaxing resource balancing for certain units and ending the practice of not making metal producing structures when metal is already full. This is to bring into play the new rule of making factories whenever there is 90% metal more often.
The bombers don't work very good just yet as they keep bombing whatever it was long after it's dead.
CPU Peak: I still don't know for sure why, but I suspect that it is related to searching for a building location, as it tends to crop up when they are not available. I'm considering doing this kind of work in a seperate thread so that it doesn't impact performance so much.
This also might be because there are too many metal spots, as on wide open combat, or that asteroid one (what's it called?).
Support for FF: I've downloaded the mod, but I haven't really played with it yet. I'll get some support for it, but I'm not that great a player, so I can't promise much.
Also, there is a bug when the first unit of a new Attack Team is a gunship, the AI crashes because air units don't have move information. I'm changing this to count the move type as hover for now.
For 1.08, I'm also relaxing resource balancing for certain units and ending the practice of not making metal producing structures when metal is already full. This is to bring into play the new rule of making factories whenever there is 90% metal more often.
The bombers don't work very good just yet as they keep bombing whatever it was long after it's dead.
Bottom kicking AI
Nice one. This AI rocks. This is the first Spring AI I've played that gives me a half decent 1v1 game. A big improvement on the last version that I d/l (which itself wasn't bad).
I did JCAI vs NTAI (0.29b) vs AAI vs OTAI (LL) on Greenhaven, all playing ARM. OTAI played way better than any of the other AIs. NTAI came out second. AAI came last, but probably only because of the 0,0,0 bug (it was in the top left corner of the map) - nobody attacked JCAI for ages.
I think NTAI has taken a step back recently. It seems "smarter", but just doesn't play as effectively as it used to. In terms of economy it just didn't expand very fast, which it used to be fairly good at. It was around three times lower on metal income than OTAI throughout the game. It put attacks together better than OTAI, but there were too many Hammers, which fired HT shots all the time and so did really badly against the very mobile attack/defense which OTAI uses.
However OTAI's d-gunning technique was probably the main factor in destroying NTAI's attacks. Awesome to watch =)
Nice one
Munch
I did JCAI vs NTAI (0.29b) vs AAI vs OTAI (LL) on Greenhaven, all playing ARM. OTAI played way better than any of the other AIs. NTAI came out second. AAI came last, but probably only because of the 0,0,0 bug (it was in the top left corner of the map) - nobody attacked JCAI for ages.
I think NTAI has taken a step back recently. It seems "smarter", but just doesn't play as effectively as it used to. In terms of economy it just didn't expand very fast, which it used to be fairly good at. It was around three times lower on metal income than OTAI throughout the game. It put attacks together better than OTAI, but there were too many Hammers, which fired HT shots all the time and so did really badly against the very mobile attack/defense which OTAI uses.
However OTAI's d-gunning technique was probably the main factor in destroying NTAI's attacks. Awesome to watch =)
Nice one
Munch
No wonder, the Arm buildtree of NTAI hasnt bene updated since some version of 0.27 when I added Xect v Mynn a TLL buildtree and the shipyard to core for basic water support..
As for economy, I had that thought too, so I got rid of a lot of the missile tower building in the buildtree, removed the lvl 2 lab from the commanders buildtree and put a lot mroe mexes and so on, in hopes of getting a 0.28 style build tree going again.
Could we see a shot of the minimap at some pivotal moment that sorta summarizes how the game went....
And maybe a redo with XTA Core instead of Arm, or 2 of each AI for core and arm allied together....
As for economy, I had that thought too, so I got rid of a lot of the missile tower building in the buildtree, removed the lvl 2 lab from the commanders buildtree and put a lot mroe mexes and so on, in hopes of getting a 0.28 style build tree going again.
Could we see a shot of the minimap at some pivotal moment that sorta summarizes how the game went....
And maybe a redo with XTA Core instead of Arm, or 2 of each AI for core and arm allied together....
yeah OTAI is really kicking major ass from what ive seen... Hats off!
I didnt have much time to test it though, does the dgun code predict its target, avoid friendlies and other fancy stuff???
Im jealous! KAIs legs and arms have been chopped off atm(to be replaced with neat code), so all it does now is squirm looking around in despair :D
Keep it up!
I didnt have much time to test it though, does the dgun code predict its target, avoid friendlies and other fancy stuff???
Im jealous! KAIs legs and arms have been chopped off atm(to be replaced with neat code), so all it does now is squirm looking around in despair :D
Keep it up!
oh well i was very curious lol, couldnt help myself in not trying it out!
well first of all it wipes my ai out very fast!
it has a very good attack system and very good nanostall control , does a (currently in the ai scenario) very nice rush defence allthough not good enough to survive a good player (wich is pretty hard->you shouldnt worry much about this, after all even vets can be raided out with a zipper/jeffy in xta)(Note: i confess it was not easy rushing it it kept building on very fast).
It is WAY WAY stronger than the last one available
!
i like pretty much the lvl1 swarm approach.
thats a very good work you did
well first of all it wipes my ai out very fast!
it has a very good attack system and very good nanostall control , does a (currently in the ai scenario) very nice rush defence allthough not good enough to survive a good player (wich is pretty hard->you shouldnt worry much about this, after all even vets can be raided out with a zipper/jeffy in xta)(Note: i confess it was not easy rushing it it kept building on very fast).
It is WAY WAY stronger than the last one available
!i like pretty much the lvl1 swarm approach.
thats a very good work you did
AI FFA Screenshots
OK, sure I'll do a rematch with core at some point.AF wrote:No wonder, the Arm buildtree of NTAI hasnt bene updated since some version of 0.27 when I added Xect v Mynn a TLL buildtree and the shipyard to core for basic water support....
Could we see a shot of the minimap at some pivotal moment that sorta summarizes how the game went....
I've captured 8 screenshots of the minimap from the game, and uploaded them here. Only problem is... I don't know where they've been uploaded to! Otherwise I'd link 'em in here. I'm a bit annoyed with myself 'cause I've spent over an hour getting the screen shots done and uploading them - it takes ages on a 56k link. The final screenshot of the eight is taken very shortly before the end of the replay - the game crashed, but I think you'll agree (if you see the screen shots) that OTAI is the clear winner.
Please - if you know where files go when they are uploaded here, let me know! Thanks. (FU upload hasn't worked for me for months - I think it has problems with the sloooow speed of my link)
Cheers
Munch
Blue = AAI
Red = OTA "LL"
Yellow = NTAI 0.29b
Green = JCAI
EDIT: I gave FU one more shot and it worked - phew =)
After the first few minutes it's fairly even:
10 minutes in and OTAI is already starting to forge a lead:
20 minutes in and OTAI has a large base in the bottom right, bringing in 25 metal, and considerable military presence in the top left. NTAI has also expanded fairly nicely in the bottom left, but on a smaller scale.
Five minutes later and OTAI now has an income of around 35 metal compared to 7 or 8 for all other players.
Another 90 seconds later and the income has jumped up to 43. We can also see one of JCAIs attack packages in the top left corner. JCAI sent a decent amount of units in that direction over the course of the game and did quite a lot of the damage to AAIs base. It also took on some of OTAIs attack packages, so that you ended up with AT and JC duking it out just outside AAIs base =)
6 minutes later NTAI (who has been left alone more or less) sends a serious assault force into OTAI's territory. There have been a number of smaller packages sent that way before, but the decent sized ones got shredded by OTAI's dgun.
Five minutes later and the NTAI attack has been neutralised. The main problems with the attack were 1. Almost all hammers 2. High trajectory fire against mobile units 3. units stayed stationary during combat, making easy targets. 4. units stayed to fight enemy troops instead of pressing in to destroy base. If these hammers had kept going and wandered round OTAI's base for a bit they could have made a serious dent. However by the end of the battle we see that OTAI's metal income has gone up another 15 to +58.
Four minutes later:final shot just before the orginal game crashed. Here we see that OTA changes it's focus to JCAI's base and flattens it fairly quickly by sheer weight of numbers. You can also see the large number of units in the top left of the map due to the afore mentioned bug.
That's all folks!
Yay!
Now onto the next version (sort of). These links are for a new multi-threaded version of OTAI that should hopefully solve the whole freeze-up once and for all, after I finish getting them debugged. These are just for test purposes ONLY and are likely to cause crashes and multithread mayhem. Source DLLs
I'm new to this whole multithreading thing so try to bear with me here. I should get out the real thing before too long, as I don't have a particualarily complicated type of multithreading going on.
Now onto the next version (sort of). These links are for a new multi-threaded version of OTAI that should hopefully solve the whole freeze-up once and for all, after I finish getting them debugged. These are just for test purposes ONLY and are likely to cause crashes and multithread mayhem. Source DLLs
I'm new to this whole multithreading thing so try to bear with me here. I should get out the real thing before too long, as I don't have a particualarily complicated type of multithreading going on.
OTAI wins with Core too
Hi,
OK so I'm a sad individual and I spent the evening running more tests - all on Greenhaven. This time with core (as requested). The main differences were 1. JCAI didn't seem to want to play ball with Core, so I reverted it to Arm, 2. AAI is much stronger with Core.
I ran the tests with OTAI starting in the top left corner to try and get round the "send units to top left" bug. However, it turns out AAI also suffers from this - it's bombers and gunships were absolutely obliterating the very top left pixel of the map.
The end result was that OTAI still won all the time, though by a narrower margin, since AAI packed a punch. In the end the constant roaming units that OTAI sends out worry away at everybody else's outlying metal extractors, and give it the edge over AAI, which was otherwise almost keeping up with it at times.
Having looked at NTAI fairly closely I think the main reason it has gone downhill is that it is more or less permanently nano-stalled. It tries to build level 2 labs before it has the income to do so, which cripples it's progress. That's my opinion based on the few games I've watched it play, noting that that's solely on Greenhaven.
A couple more screenshots below for you.
Cheers
Munch
OTAI wins with core
In another core game, OTAI takes a commanding lead (OK the tanks are outnumbered, but check the minimap)
OK so I'm a sad individual and I spent the evening running more tests - all on Greenhaven. This time with core (as requested). The main differences were 1. JCAI didn't seem to want to play ball with Core, so I reverted it to Arm, 2. AAI is much stronger with Core.
I ran the tests with OTAI starting in the top left corner to try and get round the "send units to top left" bug. However, it turns out AAI also suffers from this - it's bombers and gunships were absolutely obliterating the very top left pixel of the map.
The end result was that OTAI still won all the time, though by a narrower margin, since AAI packed a punch. In the end the constant roaming units that OTAI sends out worry away at everybody else's outlying metal extractors, and give it the edge over AAI, which was otherwise almost keeping up with it at times.
Having looked at NTAI fairly closely I think the main reason it has gone downhill is that it is more or less permanently nano-stalled. It tries to build level 2 labs before it has the income to do so, which cripples it's progress. That's my opinion based on the few games I've watched it play, noting that that's solely on Greenhaven.
A couple more screenshots below for you.
Cheers
Munch
OTAI wins with core
In another core game, OTAI takes a commanding lead (OK the tanks are outnumbered, but check the minimap)
please dont waste your time with FFAs...
If you want meaningful results, just face of each pair of AIs.
We have 4 ais -> 6 pairings
*2 to switch starting positions
*4 for each core/arm combination
=> 48 runs should give a nice gauge of current ai quality.
And then it will be all meaningless because just changing to a map a BIT different cant make a world of difference to an ai that cannot really think and adapt...
Going from painted dessert to comet catcher can make the difference between ai a wins everytime to it losing...
If you want meaningful results, just face of each pair of AIs.
We have 4 ais -> 6 pairings
*2 to switch starting positions
*4 for each core/arm combination
=> 48 runs should give a nice gauge of current ai quality.
And then it will be all meaningless because just changing to a map a BIT different cant make a world of difference to an ai that cannot really think and adapt...
Going from painted dessert to comet catcher can make the difference between ai a wins everytime to it losing...
4 way ffa
HI There,
I agree that if you just want to find out whether one AI is better than another then 1v1 is probably better. However, I don't have time to run 48 tests my self and I find watching 1v1 games fairly boring. Also 1v1 games require very little recon on the part of the players because you can get away with just throwing your units at the opposite corner of the map most of the time.
Playing in a 4 way ffa is a different skill to 1v1, as I'm sure you yourself know from playing against other humans... and like I say it's certainly more fun to watch =)
Historically Greenhaven was one of the main testing grounds for AIs. In OTA, if your AI did well on Greenhaven then it probably did OK on most any land map - that's why I chose it. Well that and the fact it looks nice =) Of course with Spring maps being very different, Greenhaven may no longer be such a good test map, but it will do for me =)
Go ahead and do some more tests yourself and post some shots. Maybe we should start a separate thread though.
Cheers
Munch
I agree that if you just want to find out whether one AI is better than another then 1v1 is probably better. However, I don't have time to run 48 tests my self and I find watching 1v1 games fairly boring. Also 1v1 games require very little recon on the part of the players because you can get away with just throwing your units at the opposite corner of the map most of the time.
Playing in a 4 way ffa is a different skill to 1v1, as I'm sure you yourself know from playing against other humans... and like I say it's certainly more fun to watch =)
Historically Greenhaven was one of the main testing grounds for AIs. In OTA, if your AI did well on Greenhaven then it probably did OK on most any land map - that's why I chose it. Well that and the fact it looks nice =) Of course with Spring maps being very different, Greenhaven may no longer be such a good test map, but it will do for me =)
Go ahead and do some more tests yourself and post some shots. Maybe we should start a separate thread though.
Cheers
Munch
What i wanted to say is
a) that one game is meaningless
b) ffa really can warp the results.
ALso, things like core or arm can make a HUGE difference.
I personally only test otai and aai (as the other two have crashed once too often on me), and the results are really unpredictable between maps and teams.
ALso, especially with AAI you have to consider the status of the learning file for the mod and the map.
If it was on the same start position before, it will for example build defensed in the direction it was attacked the last time, which will warp result if there are many runs.
Also, i would suggest waiting for the next aai version, which is quite a bit stronger than the current one.
Also i noticed that AAI sucks at XTA (which i think is sad, as i think AA is for hyperactive kids...
There are quite a lot of factors.
Btw: does OTAI run on ALL maps now without crashing?
a) that one game is meaningless
b) ffa really can warp the results.
ALso, things like core or arm can make a HUGE difference.
I personally only test otai and aai (as the other two have crashed once too often on me), and the results are really unpredictable between maps and teams.
ALso, especially with AAI you have to consider the status of the learning file for the mod and the map.
If it was on the same start position before, it will for example build defensed in the direction it was attacked the last time, which will warp result if there are many runs.
Also, i would suggest waiting for the next aai version, which is quite a bit stronger than the current one.
Also i noticed that AAI sucks at XTA (which i think is sad, as i think AA is for hyperactive kids
...There are quite a lot of factors.
Btw: does OTAI run on ALL maps now without crashing?
hmm, I am sure that nanostall due to lvl 2 plants was less likely to happen with 0.29b because the commander no longer attempted to build the lvl 2 kbot plant, however I dont know how different the metal distributions are on that map, having always tested on maps such as comet catcher remake, the cold place, altored divide....
New half-version of OTAI. Now at a full 1.08. The threads don't seem to crash the game so far, but they do even out the CPU peaks into more of a plateau. Instead of locking up, the games gets slower.
OTAI should play OK on all maps, as long as it uses the right profile. I've carved them down to Default for most maps, and Gunship for maps where you can't walk over to the enemy (Blue Planet, for instance, or Valles Marineris or some Island Map). I'll add hover support later.
I would like to agree with IMSabbel that doing FFA does warp the results somewhat. I mean, if JCAI and NTAI both attack AAI at the same time, and them OTAI picks up the pieces, does that really mean OTAI is better, or that AAI was just unlucky or in the top-left corner? I personally like to watch FFAs better, but I think 1vs1 is what counts.
@IMSabbel
OTAI should work on all maps without crashing. I can't promise anything if the map lacks metal or lacks building spots near the start points. Should work on all 'normal' maps and metal maps.
Oh, and I moved the download position to fileuniverse. See the first post for the site. I forgot to bring the source code this time, but it shouldn't have changed too much.
OTAI should play OK on all maps, as long as it uses the right profile. I've carved them down to Default for most maps, and Gunship for maps where you can't walk over to the enemy (Blue Planet, for instance, or Valles Marineris or some Island Map). I'll add hover support later.
I would like to agree with IMSabbel that doing FFA does warp the results somewhat. I mean, if JCAI and NTAI both attack AAI at the same time, and them OTAI picks up the pieces, does that really mean OTAI is better, or that AAI was just unlucky or in the top-left corner? I personally like to watch FFAs better, but I think 1vs1 is what counts.
@IMSabbel
OTAI should work on all maps without crashing. I can't promise anything if the map lacks metal or lacks building spots near the start points. Should work on all 'normal' maps and metal maps.
Oh, and I moved the download position to fileuniverse. See the first post for the site. I forgot to bring the source code this time, but it shouldn't have changed too much.