23 replies to this topic

[LarkinVB]

2.50b2 is up for the testers!

Changes
======
- Added emergency retreat for sorceror and mek boy
- Improved Necron Warrior spamming routine to force 3 NW squads with 1 monolith and 4 NW squads with 2 monoliths
- Some more speed optimizations (slight ones)
- Improved attack gathering behaviour
- Increased req donation limit to 1000

Larkin wrote:

I've heard v2.5 is to be released soon. Anything new besides speed increase ?

I'd suggest adding EmergencyRetreat for big mek and sorceror too.

[ArkhanTheBlack]

I've heard v2.5 is to be released soon. Anything new besides speed increase ?

I fixed the Tau weapon stance bug you mentioned, but that's all besides the speed fixes. Originally, I had no plans for a 2.5 revision, since 2.4 is pretty much bug free and already contains most AI improvements I wanted to be included. I started 2.5 because of the player reports for the mods which are using our AI. If a player is playing vanilla DC with a slow PC he can just decided to use Relics DC AI, but if they are playing DOWPro, DOWXP or ID they can't just switch our AI mod off if the have performance problems since there's no other AI available.
After my 2.2 optimisations, the AI Highspeed option lost a lot of its effiency compared to the full AI, and I therefore wanted to make this option a usefull alternative again. It should be now more than twice as fast as the full AI and I hope that'll be enough to please even players with low end PC's. I also saw two ways to improve the basic AI speed quite a bit and the changes were finished quicker than I thought.
2.5 is only intended as a mini revision for speed improvements, but if you have some usefull changes which don't take too much time, I can add them. The emergency retreat for sorcerors and big meks sounds good. Usefull changes without much work...

[Inquisitor]

AI Version: 2.5b1
Matches: 1v1 three times
Factions: Necron AI vs IG AI
Game Modes Used: Annihilate
Map: Various
AI Difficulty: Harder
Resources Rate: Normal
Resources Sharing: Off
Winner: IG every time

Comments: I took 2.5b1 for a swing and I wanted to see how the AI Necron was playing. I figured I'd give them an easy match and pitted them against IG. Well, IG just wiped the floor with living metal in each game!

Observations of interest: Necrons were helpless against the AI IG, even though the NL is played quite well using SP and Phylactery, and other powers. I think that this is contrary to what most DC players would expect and something must be done to improve the way the AI plays Necrons. Monolith(s) were idle when there was free pop. I only saw one FO squad in the three games.

Suggestions: Bearing in mind what ArkhanTheBlack said about the purpose of this release, my only suggestion would be to reduce the idleness of the Necron Monolith, if at all possible. It might help dramatically if the AI built more "free" warriors. I know that the third/fourth warrior squad takes forever to build and that it might even hurt the AI, but if it is easy to do I could test it thoroughly within short notice.

Replays: Available, but two of them were played on custom maps, so these will have to be installed as well.

[thudo]

1) As a tester I would dissuade you from running AI vs AI.. we can do that ourselves as devs. You shouldn't. BetaTesting demands the HUMAN FACTOR to trick our AI and, thus, allow us to script more adaptive opponents. Having AI vs AI is NOT a good test result per se. It is somewhat, but generally we need your own battlefield experience as a hardened DoW player.

2) Wow.. Necrons defeated by IG? What the..?!?! Surely this isn't due to 2.50 as Necrons rule almost all games I find. I'll check and see what gives.

Thanks Inquisitor for helping us make the best DoW AI possible!

[Inquisitor]

1) As a tester I would dissuade you from running AI vs AI.. we can do that ourselves as devs. You shouldn't. BetaTesting demands the HUMAN FACTOR to trick our AI and, thus, allow us to script more adaptive opponents. Having AI vs AI is NOT a good test result per se. It is somewhat, but generally we need your own battlefield experience as a hardened DoW player.

2) Wow.. Necrons defeated by IG? What the..?!?! Surely this isn't due to 2.50 as Necrons rule almost all games I find. I'll check and see what gives.

Yes, I know I shouldn't be posting results with AI vs AI, but when IG won 3 of 3 games I figured that I had to report it. I never thought that it would be possible to code the IG AI to consistently win vs. Necrons, since it requires almost insane microing skill for a human IG to win vs. a non-noob Necron in DC1.2! Essentially it illustrates how good a job you're doing with the AI! So kudos to the AI team for pushing the limits of the AI even further!

Anyway, I can't help thinking that the Necron AI must be a real easy opponent in DCPro or DoWXP (especially) given the rebalancing in those mods...

Btw, I previously played a game as Necrons vs. SM AI on Battle Marshes on Harder and it annihilated me (no replay available). I'll do a rematch today and see if I can do better. I probably just need to clear my head and concentrate to play a better game. I'll post some results when they're available.

[Maktaka]

I've got a question about how you guys handle ally support. I've noticed that the AI has a tendency to move, rather than attack move, supporting troops through the attacking forces to defend their ally, even on clear shots through the enemy force (such as on Into the Breach). Needless to say, this makes for a bunch of dead reinforcements. If I remember correctly, the decision to use move or attack move when sending reinforcements is dependent on a Relic function to determine where the attackers are. Is that still the case, or is there something you have control over that could improve behavior?

Also, a couple minor suggestions. In DS 2.4, the AI will regularly start the game by donating resources to me, which seems rather counter-productive for them when resources are so tight early game. Perhaps a delay could be added to prevent resource donations during the first 30 seconds, or maybe just slightly increasing the amount of resources needed to trigger a donation.

I also remember some reports during the 2.4 beta of Necrons building forward power generators. Unfortunately, this quirk seems to have survived through the beta because I still see forward generators near, say, the Relic at the entrance point of the bases in Mortalis. Although the Necron AI is smart enough to build them on the close side of the Obelisk, they shouldn't build them there at all as long as spare room is available around the rear Obelisks and Monolith.

Other than those minor quibbles, this is still the most regularly satisfying way to play the game. Great jorb! Amazing how far you guys have come since the early releases.

[LarkinVB]

Nec vs IG . Nec won. IG did float a lot of req (2000) short before T3. Squads were few and small but it did not spend req on more or bigger ones. It wasted much of this req on a secondary HQ when T3 was finished and lost the game because it had not sufficient infantry.

[Inquisitor]

Nec vs IG . Nec won. IG did float a lot of req (2000) short before T3. Squads were few and small but it did not spend req on more or bigger ones. It wasted much of this req on a secondary HQ when T3 was finished and lost the game because it had not sufficient infantry.

Weird. Reading your post, I just did another game on Blood River. An epic battle that lasted 25 min (game time) ended with the Baneblade finishing the job.

Replay

[LarkinVB]

Nec vs IG on titan fall (first game described above was on meeting of minds). Nec won easily though the game was long. It's wasn't long because of hot fights but because of minutes of complete inactivity midgame.

Nec did build three monos but none of them engaged. Bug ?

Can you have a look on the gather code, Arkhan. I find that armies often attack peacemeal, even if squads have the same speed.
This was asked for in the dcpro forum too.

Attached Files

•  2P_TITAN_FALL.2007_07_28.12_45_55.zip   208.1KB   16 downloads

Edited by LarkinVB, 28 July 2007 - 10:59 AM.

[Inquisitor]

Nec vs IG on titan fall (first game described above was on meeting of minds). Nec won easily though the game was long.

Even weirder. Just did the same match and IG was again vastly superior.

Can this have anything to do with different machine specs? I run on a reasonably fast system, Core 2 Duo @ 3.6 and a 8800 GTS.
EDIT: On Vista32 with 2GB memory. Note that attachment is a renamed 7zip file (due to attachment rules).

EDIT2: IMPORTANT: I get a sync error when I try to play your replay. Could anybody with 2.5b1 please check get they get a sync error with mine (mine's definitely recorded with 2.5b1). If you do, I might have something wrong in my DC installation (I have tried a lot of mods recently, so it's not impossible that it's me).

Attached Files

•  2P_TITAN_FALL.2007_07_28.14_57_49.7z.rar   117.7KB   13 downloads

Edited by Inquisitor, 28 July 2007 - 01:17 PM.

[ArkhanTheBlack]

Can this have anything to do with different machine specs? I run on a reasonably fast system, Core 2 Duo @ 3.6 and a 8800 GTS.

Highly unlikely. I guess you are using different settings like Larkin (Heroes, difficulty, etc.), or it's just the choice of the start strategies.

Nec did build three monos but none of them engaged. Bug?

Mono was always bugged in several ways (Engaged building, repairing, etc.).

Can you have a look on the gather code, Arkhan. I find that armies often attack peacemeal, even if squads have the same speed.
This was asked for in the dcpro forum too.

I've removed the doubled gathering range after 60 seconds. They'll now add 10 range after 60 sec, 20 after 90 and 30 after 120. We'll see if it works.

Nec vs IG . Nec won. IG did float a lot of req (2000) short before T3. Squads were few and small but it did not spend req on more or bigger ones.

Units which don't cost power are now allowed to reinforce if req is greater than 800.

I've got a question about how you guys handle ally support.

This is a very old story. We use a radius check for the attack/normal move decision which is flawed. A good solution would require a complex battlefield analysis of where a squad is moving to and what enemy units are on the path. We can't do that since we don't have the necessary pathing functions.

Also, a couple minor suggestions. In DS 2.4, the AI will regularly start the game by donating resources to me, which seems rather counter-productive for them when resources are so tight early game. Perhaps a delay could be added to prevent resource donations during the first 30 seconds, or maybe just slightly increasing the amount of resources needed to trigger a donation.

I've increased the donation limit to 1000 req. I guess this should work.

Suggestions: Bearing in mind what ArkhanTheBlack said about the purpose of this release, my only suggestion would be to reduce the idleness of the Necron Monolith, if at all possible. It might help dramatically if the AI built more "free" warriors. I know that the third/fourth warrior squad takes forever to build and that it might even hurt the AI, but if it is easy to do I could test it thoroughly within short notice.

They should force 3 NW squads with 3 monos and 4 NW squads with 2 monos. We'll see if it works..

[Inquisitor]

I guess you are using different settings like Larkin (Heroes, difficulty, etc.), or it's just the choice of the start strategies.

But knowing how IG AI usually fare against Necron AI, I think it is a bit suspicious that they win 5/5 games on my machine and lose 2/2 on LarkinVB's...

All my games were played with the settings mentioned in my report a few posts back. I had Necrons as player 1, LarkinVB had them as player 2 if that could make a difference...

They should force 3 NW squads with 3 monos and 4 NW squads with 2 monos. We'll see if it works..

I guess you mean 4NW with 3 monos and 3NW with two... I'll be sure to test it as soon as it is released!

[LarkinVB]

I think the difficulty level might be the difference. I do all my tests with hard AI, no heroes. Here is a game Nec vs IG on blood river. IG stood no chance as expected. IG played very poorly, just running around in circles through the enemy. Perhaps you can have a look Arkhan. IG completely failed to gather it's squads and shooting it out. The preferred running, obviously not in attack move.

EDIT2: IMPORTANT: I get a sync error when I try to play your replay. Could anybody with 2.5b1 please check get they get a sync error with mine (mine's definitely recorded with 2.5b1). If you do, I might have something wrong in my DC installation (I have tried a lot of mods recently, so it's not impossible that it's me).

I get a sync error with yours.

Attached Files

•  2P_BLOOD_RIVER.2007_07_28.20_21_46.zip   156.53KB   11 downloads

Edited by LarkinVB, 28 July 2007 - 06:37 PM.

[thudo]

v2.50b2 is up for da testers!

Changes
======
- Added emergency retreat for sorceror and mek boy
- Improved Necron Warrior spamming routine to force 3 NW squads with 1 monolith and 4 NW squads with 2 monoliths
- Some more speed optimizations (slight ones)
- Improved attack gathering behaviour
- Increased req donation limit to 1000

[ArkhanTheBlack]

I think the difficulty level might be the difference. I do all my tests with hard AI, no heroes. Here is a game Nec vs IG on blood river. IG stood no chance as expected. IG played very poorly, just running around in circles through the enemy. Perhaps you can have a look Arkhan. IG completely failed to gather it's squads and shooting it out. The preferred running, obviously not in attack move.

That was one the nastiest type of bug I know. If I wouldn't have seen the phenomenon before, I highly doubt that I would have found the cause so soon. In my own test game, IG made a crazy attack on th Necron base while their own HQ was under attack. Such a behaviour is normally not possible for the AI. That combined with the crazy movements was a strong indicator that they had gotten an invalid position outside the map. And indeed, the friendly base array had stored pointers to the post positions instead of vector copies of the positions (I HATE this unclean and confusing LUA behaviour!!!). The result were 'moving' post and base locations. Don't ask me why they are moving, because I have no idea, especially not why it's only happening on some maps. But since I had this bug before, I at least knew what happened. As soon as a target position moved off the map, the units went crazy.

Edit: It's not fixed in beta 2 yet. You have to wait until beta 3 for that.

Edited by ArkhanTheBlack, 28 July 2007 - 11:49 PM.

[RobertW7928]

i don't know if you noticed but in v2.4 online with Ai donations on you get sync errors, have you noticed or got it?, i was hoping for a fix?

[Inquisitor]

I get a sync error with yours.

I found that I could not view replays from dow sanctuary as well, so something is up with my DC installtion. I have recently installed DoWXP, DCPro, ID, Tyranid mod and one of them must have changed something in DC1.2.

I have just now uninstalled DC, reinstalled (to the same location without deleting anything), patched and tried to view a replay from dow sanctuary and I still get a sync error. Anybody have any suggestions besides deleting all files before reinstalling? I would rather not, since I would have to reinstall all the custom maps and mappacks and backup my profiles etc...

Edited by Inquisitor, 29 July 2007 - 06:26 AM.

[LarkinVB]

I'll wait for b3.

[Inquisitor]

Alright, I think I have a major find which might be related to the positioning bug that ArkhanTheBlack mentioned.

I have reinstalled DC to another location and I can now play LarkinVB's replay. (I would still like suggestions to what could cause sync errors - even after reinstall!)

Well to get back to the find, I played three games with Necrons in position 1 and three games with IG in position 1 as well, and guess what: IG won all games on position 2 and Necrons won all games on position 2!

AI Version: 2.5b1 (to rule out any changes in b2)
Matches: 1v1 six times
Factions: Necron AI vs IG AI on different locations
Game Modes Used: Annihilate
Map: Titan's Fall
AI Difficulty: Harder
Resources Rate: Normal
Resources Sharing: Off
Starting Location: Fixed
Winner: IG 3/3 on pos 2 and Necrons 3/3 on pos 2

Comments: I was interested in finding the reason for the IG AI winning every game vs. Necrons.

Observations of interest: IG won all games on position 2 and Necrons won all games on position 2.

Replays: Six replays

Edited by Inquisitor, 29 July 2007 - 05:00 PM.

[thudo]

IG won all games on position 2 and Necrons won all games on position 2.

Explaination: badly designed map. If a side ALWAYS consistently wins based on its starting position that means the map is grossly bias towards a startpoint. Seen this too many times but ohh well.. you can't have perfectly symmetrical maps or else life would be bloody boring.