76 replies to this topic

[thudo]

Go get em boy-os!

Point-form your comments and best tested on HARD+ on either 1vs1 or 2vs2 maps. Try 2-3 different maps and thanks!

[ThetaOrion]

Well, I have deadlines at work, so I'm out of the Beta-testing thing for the most part. I don't have time to study the replays.

But, after seeing nobody respond here for over a day, I decided to give it a go with Beta 11. Thudo invited me to give it a look.

Mountain Trails 2vs2, HARD setting, me Chaos 1, Orks AI Ally 2, against IG 3, and SM 4.

Well, that was definitely different. I won! That was the first time I have won that scenario without -dev in about the last five or six times I have tried it without -dev. All previous times I lost big or hit stalemate without -dev.

This time, the Ork AI ally was great!! They moved out and kept up the pressure the whole game. Wonderful!!

The IG AI enemy didn't build any Kasyrkin or Ogryn, which would have helped them a lot, and it would have helped the IG if they would have attached their Priests to some of those Guardsmen squads. But, the IG built a lot of tanks, which gave me grief, so it all balances out in my opinion.

1) But, look at the replay, for the main flaw in the AI that I noticed during this particular skirmish.

The SM AI enemy took their whole army over to their enemy's Critical Point there on the west and 'guarded' the Ork's captured critical point. The Space Marines got stuck there, and didn't have the power to decap or capture the critical point. They just picnicked or got lost or just stood there and vibrated. It was a deadend in their logic. They remained stuck even while their base was being destroyed, and no further plans freed them to move on to defend their base or to decap and capture the critical point.

It's always pleasing to win, especially after losing that scenario a half dozen times in a row. In those Beta 7 and Beta 8 scenarios, the Space Marines were ultra uber, and the Orks were weak. This time in Beta 11, the Orks were great, and the Space Marines had gathering problems and didn't seem to be building turrets either around their start position. I just remembered SM turrets around a listening post, anyway the replay will get you what I failed to properly remember.

There must be a hidden directive hardcoded in the 1.41 WA core that states that the Orks and the Space Marines cannot be simultaneously great, at least on the Mountain Trails Map.

2) You'll have to watch the replay to see what else I might have overlooked, either good or bad. But, at least there is now a Beta 11 replay here with a human in the mix.

The only flaw with the Mountain Trails map is that it doesn't have enough resource points and slag deposits, so you often don't get to see how the AI performs over the long haul in a very long game. The limited resources and maybe the small size of the map can prevent some of the factions from completely building and tiering up. But, it's always an interesting fight nonetheless.

Edited by ThetaOrion, 09 March 2006 - 05:20 AM.

[ArkhanTheBlack]

Thanks that you took the time Theta!

IG tried a fast tech, and so SM got a hard stand at the start in the fight for map control. The IG fast tech succeeded, but the lack of a heavy melee walker made them a bit helpless against the chaos berserker squads. In addition they suffered heavy losses against the rocket turrets.
SM couldn't capture the critical since they had not even one capturer in their attack force. Bad luck! They had to wait until a new squad is built which is able to capture. There was a single scout capturing the strat before. After that he might have tried the crit. It was too late anyway. BTW, SM built several turrets.
In my opinion IG lost the game not SM.
But so far I didn't see any bigger AI issues except the known ones like the melee probs.

[Quietdeath]

Found a little Bug:

The eldar begins to build a soul shrine but does not finish it - I do not exactly know why - maybe because the builder was under fire.
Here the Replay:
Battle Marshes: SM v Eldar

I also played two games again Orcs and I really like the way they play. In Fallen City I attacked his base and he invaded my base with killa cans
Although the jump code is pretty good, I had one issue while playing. The stormboys jumped away in the right moment but towards my base and so they got killed.

[thudo]

Yes. I did notice the occassional builder not finishing or starting off on a building. Odd. Its semi-rare. It has been known that builders will sit around while some buildings sit there and are not finished construction. If the base where the idle builders are is assaulted then the builders' code will "kick in" and all is back to normal. Again.. its somewhat a rare phenomenon.

About the enemy melee jumpers jumping to your base thus wrong direction.. well.. heheh.. move that base somewhere else damn it! Guess that involves extra coding to avoid enemy bases now. Ugh!

[ThetaOrion]

The Trends after my my Absence:

1) Back in Beta 7 and Beta 8, it was as if my AI enemies were locked into the exact same build programs one game after another without -dev on that same Chaos 1, Ork 2, IG 3, SM 4 scenario. It played out the exact same way every game for like five or six games in a row, even though I as the human tried to do things differently. I lost or hit stalemate every time, and the Space Marines were ultra uber and the Orks were weak. I could never seem to get a different set of build programs. It was as if the AI randomizer wasn't even working.

2) This single Beta 11 game was like no set of build programs I have seen before. For thinking that the Orks couldn't be improved back in Beta 7 or Beta 8, the Orks AI allies were really really wonderful during this Beta 11 game. The Orks won the game as far as I am concerned. You guys have done a really bang up wonderful job with the Orks!! The IG were fine, too, they did what they do, and they gave me a lot of grief.

3) In this Beta 11 game, it would have been nice if the Space Marines would have abandoned the critical point and gone back to defend their base, but it was late in the game and they were going to lose anyway, just like you said Arkhan. You know, after two months of doing this beta-testing, I get the feeling that the Space Marines were never really programmed to be on the losing side. Once the Space Marines are on the losing side of resources or the losing side of the game, it's as if they just don't know what to do. The Space Marines seem to only know what to do and do it well when they are on the winning side of resources and the winning side of the game. This is just an overall trend I'm noticing going on ever since the introduction of 1.41 of Plain Vanilla Winter Assault.

4) Back in Beta 8, the Eldar literally started ten different turrets around each listening post but didn't finish a single one of them, on one of the games I played. This was a large map, so the Eldar had plenty of time to build unmolested, because they were surrounded and protected by their AI allies. However, while everyone was building Ubers or Titans, the Eldar didn't have any Fire Prisms or nothing except a couple of troops and a WraithLord when I finally got to them in Tier 4. All of their resources had gone into building started-but-unfinished turrets. The Eldar sometimes get off track and don't finish building what they start. Maybe the problem is still there, per QuietDeath's report? I found, though, that to get the Eldar's start-but-not-finished problem to be really magnified, it needs a large 8 player map and a long time of their being off track for it to become really noticeable. This off track thing is limited to one unfinished turret or one unfinished Soul Shrine on the smaller maps. It's the big maps and long games where the Eldar have a long time with their base untouched by enemy fire that these fields of unfinished buildings become really noticeable and hurts the Eldar's performance. It's as if some far distant on the other side of the map Eldar troop taking fire is enough to divert the Eldar builders back home from their task of building. I don't understand it. But, there's a possibility that that Eldar bug is still there. Need somebody with the time to try a large 8 player map and put the Eldar in a protected spot and leave the Eldar base unmolested for the whole game and see if they get off track in a big way.

Sorry, but I have got to get back to work, no time to try another game with a human in the mix.

[ArkhanTheBlack]

Builder bug happens from time to time. I think it's a hardcoded bug from Relic.

Actually the jump code itself is innocent. The guilty part is the attack plan. Fixing that is a hard one...

Kind of funny that everyone gives a positive feedback about orks. It seems like they did profit a lot from the changes in the movement code, since I didn't change the orks at all.

[ThetaOrion]

As I understand it, this Eldar build bug doesn't happen all the time, even now. And, the first time I ever saw the Eldar build bug was Beta 8, and I had played dozens of games over two months of time with 1.60 Release and Beta 1 through Beta 4 on these huge unapproved 8 player maps that I have been betatesting for Aralez, and I never once saw the Eldar build bug during that whole time. We as a group noticed that the 1.65 AI was no longer building mines back in Beta 4 or Beta 2, but I never noticed any fields of started but not finished turrets. That kind of thing is rather noticeable when you are overrunning an Eldar Base late in a long game on a huge unapproved 8 player map, and I played dozens of those huge games pre-Beta 5. I only started seeing fields of unfinished Eldar turrets back in Beta 8, or maybe Beta 7.

This all kind of leads me to believe that the problem resides in one of the Eldar build programs and it was introduced some time after Beta 4. And, I think it requires a human player in the mix, but I have no way of knowing that for sure.

Edited by ThetaOrion, 22 February 2006 - 11:51 PM.

[ThetaOrion]

--
||
--

Kind of funny that everyone gives a positive feedback about orks. It seems like they did profit a lot from the changes in the movement code, since I didn't change the orks at all.

It's amazing how minor changes can sometimes have such a major impact. I think that was the lesson LarkinVB was trying to teach me the last time we interacted on this forum.

A week ago, when Thudo said that ThetaOrion must end it now, I went back to 1.60 Official Release to betatest some of these large maps for Aralez. That was an interesting experience. Everyone had a Titan or an Uber, and there were mines, and the whole 1.60 thing had a different feel that I liked a lot.

If you get a chance, go back and play 1.60 Official Release of the AI Mod, or 1.50 of the AI Mod. That trip through nostalgia can give you a longer term trend perspective on the whole thing. You see trends that you like and trends that you don't. Human nature.

Or, better yet, if you don't want a human in the mix, have Excedrin do some more of those Cage Matches with the current beta against 1.60 of the AI Mod or against 1.50 of the AI Mod. Now, you shouldn't expect the new Beta to win every time. Such an expectation is unrealistic. But such comparisons will help you to see the trends that you like and the trends that you don't like, and then you can decide what to change or improve if anything in the light of knowledge or the light of long-term perspective.

After playing around with 1.60, I uninstalled Dawn of Heresy and the 1.5 Tau Mod, and my WA game stopped working. It would read the CD, but apparently was no longer passing the test. No error message, I just couldn't get into the game -- desktop the whole time. And, my DoW CD was also locked out. So, this past week, I uninstalled WA and DoW and uninstalled every mod, and I reinstalled everything. So, now I have a fresh install of Winter Assault and fresh AI Mod 1.60 and none of the old replays. And, I just swapped in Beta 11 last night. It's the only mod I have working right now.

Now look what you did, you guys got me started again with my commenting. Gotta go away and get some work done. The problem is that I need to use this machine to do my work, and your input keeps drawing me back in. I really am going to go away now. Watch, he's going away. Come on now, why isn't he going away? We want him to go away.

[ThetaOrion]

Okay, after the Beta 11 reboot graphics crash, I had to test and see if it was all working again.

Decided to get a mix with the Eldar AI on the map, since I was going to test anyway.

Mountain Trails HARD setting Beta 11, me as SM 1, Eldar AI ally 2, against Chaos 3 and Orks 4, another good vs evil.

The Eldar finished everything they started as far as I can tell. They just didn't go down the unfinished buildings trail this time around, if they even are going to do so in Beta 11.

The Eldar were great AI allies.

The Orks were there. I got a sampling of grief from both the Orks and the Chaos early in the game, and did some dancing. When it got hot, the Orks huddled up or hid in their base, but the Chaos didn't go down easily. So, here's a scenario where the Ork AI wasn't exactly the star of the show, but they did okay.

There was nothing that I was dissatisfied or disappointed with on this scenario, and I won. Of course, watching the replay might bring something to your attention that I missed, so I include the replay.

Edited by ThetaOrion, 09 March 2006 - 05:21 AM.

[ThetaOrion]

It is hard to know what comprises a true trend and what does not, especially when you have a human making the observations. That’s what I found so fascinating about those Cage Matches – it was an opportunity to measure or observe the overall trends in a less human influenced subjective manner than simply feeling the changes or speculating about the changes.

Nevertheless, it is a long-held personal bias or personal belief of mine that the 1.60 Official Release of the AI Mod works perfectly fine on those huge unapproved 8 player maps that I have been beta-testing for months, so any trend away from that tends to get noticed and commented on at times. Aralez makes those maps for 1.60 of the AI Mod and tests them with 1.60 of the AI Mod. The Official 1.60 AI Mod is doing just fine with those maps as far as we can tell, but occasionally, I’ll come across a beta or a new build program that isn’t doing fine with some of those maps, the endless fields of started but not finished Eldar turrets being one example or some missing AI Uber Titan Avatars being another example. I never saw anything like that or heard anything about that being reported with the 1.60 Official Release of the AI Mod.

Some trends or changes you like, and some you are not as excited about. It’s human nature. And, we never know if mentioning it will actually help the coders or simply frustrate them. Yes, humans are subjective or biased about their observations and feelings instead of being objective. But I do believe that, 99 percent of the time, 1.70 of the AI Skirmish Mod will actually be played with a subjective biased human somewhere in the mix. And, maybe up to half the 1.70 AI Mod games of the future will actually be played on the larger unapproved maps. 1.60 of the AI Mod seems to handle these things just fine. So, any trends away from that might get noticed or might not always get you what I believe you are hoping to achieve.

And life goes on.

[LarkinVB]

Watched the replay, some observations :

1. Ork played poor. They wasted their complete army on attack while eldar took nearly all their points at home. No chance for a comeback, defense would be the right decision. Gretchins were fightinh hand to hand instead retreating.

2. Chaos didn't keep a strong force while teching. They had only 6 pop on the field but decided to build two turrets instead. Those made no impact.

3. Eldar should research FoF much faster. It was way after the 3 minute mark which is too late for initial dancing.

4. All races should try to get their grenades as fast as possible.

5. I found that an fast second builder is better than a fast capping squad.

[ThetaOrion]

While testing the final release of the DrawBridge 3vs5 Map for my friend, the one that caused me the endless fields of turrets with the Eldar back in Beta 8, I decided to make the test doubly useful. I set myself up as Chaos with Ork and Chaos AI allies, and I put 5 enemy Eldar outside. So, there were Eldar in the spot where the endless fields of turrets showed up in the past and Eldar everywhere else as well.

Then I played the game with Beta 11 for forty minutes without -dev and won.

This time, I busted open time to watch the replay. There wasn't a single thing that the Eldar builders started that they didn't finish. In fact, they sometimes worked together to finish things faster. I was very impressed. Each faction built two turrets early on, a couple of more turrets in the middle of the game, and then when I came out of the gates, they each built about four more turrets on average. Everything they started, they finished. They did wonderfully!!

So, either Beta 11 doesn't have the problem with the endless fields of started and unfinished Eldar turrets, or I lucked out and didn't get the single build program that produces the problem. Good deal all around, don't you think?

The Eldar were a bit slow getting their Fire Prisms, but I saw a couple of them at the end of the game. And, I saw two Eldar Avatar. I and my Chaos buddy got out BloodThirsters rather early and first. I didn't see a Squiggoth, but while watching the replay one of Da Mek Shops build something that took a very long time to build. But, by the time it was built nothing came out of the building because the Orks had reached their 100/20 caps by then. Anyway, I do believe that there was a Squiggoth there waiting in the wings.

I have been pretty pleased with the Beta 11's performance, but then I don't have Larkin's schooled and masterful eye for watching replays. He would see stuff that I don't see.

At least in Beta 11, the Space Marines aren't showing up in my base with a LandRaider and fully capped with multiple squads and a half dozen other tanks while I'm just starting to produce my second tank.

DrawBridge is one of those unapproved maps, so I won't try to upload the replay, but the Eldar did fine in my opinion when it came to building their bases. No huge or nasty surprises. Thought you might want to know. This just might be Beta where you will want to start doing Cage Matches and having LarkinVB watch the replays. LarkinVB can find stuff that the rest of us overlook, and that's a good thing if you are wanting 1.70 to be all that it can be.

So far, the Space Marines refusal to return and defend their base on Mountain Trails was the only thing to really disappoint me. Not bad. I do miss the mines, and I do sometimes wonder if Thudo ever got his mine workaround to work properly. Thudo said he had the fix for the mines, but I haven't seen it in action.

Edited by ThetaOrion, 23 February 2006 - 11:55 AM.

[LarkinVB]

@Arkhan

Perhaps orks should count squads + defensive buildings. If their number of CC squads is less than enemy ranged squads/buildings they should think about retreating as the can't bind all in CC.

[ArkhanTheBlack]

1. Ork played poor. They wasted their complete army on attack while eldar took nearly all their points at home. No chance for a comeback, defense would be the right decision. Gretchins were fightinh hand to hand instead retreating.

Yes, orcs tried too long, too hard and too far away from home. I've increased the range of the CheckAttackSituation() method. It was 35 which is probably too inacurate. I increased it to 50 at tier 1 to 65 in tier 4. I also removed the attack bonus for attacking bases in tier 1. They should now be a bit more restricted on LP targets they can handle in tier 1.
Gretchins normally flee. I think the GetThreatOnPath method gets sometimes confused if the enemy is mixed with the gretchins and returns a thread in every direction.

2. Chaos didn't keep a strong force while teching. They had only 6 pop on the field but decided to build two turrets instead. Those made no impact.

Actually Chaos did quiet well. They were able to defend against SM + Eldar pretty much alone for quiet a long time.
Turrets are tricky on this map! On others they are nearly a must have.

3. Eldar should research FoF much faster. It was way after the 3 minute mark which is too late for initial dancing.

When would you research it? Directly after barracks and generator? At the moment farseer and one combat squad is built before the research starts.

4. All races should try to get their grenades as fast as possible.

Chaos made a raptor rush, therefore grenades are delayed in this case.

5. I found that an fast second builder is better than a fast capping squad.

That would be builder - scout - scout - scout then.

Perhaps orks should count squads + defensive buildings. If their number of CC squads is less than enemy ranged squads/buildings they should think about retreating as the can't bind all in CC.

Almost impossible to get reliable results with that! If the turrets are at an unimportant position, they shouldn't be counted and the squad count is to inaccurate. One scout trooper counts as one ranged squad. It gets even more complicated if two or more opponents are involved.

[ArkhanTheBlack]

Ah, before I forget it... I removed mines only because I couldn't control how many are built from them. If they start once, they'll never stop. If you consider 7 AI's and each building 100 - 200 mines, you can get 1000 - 2000 mines on the map. Since I can't guaranty stability with such insane mine counts, I decided to remove them. Better no mines and a stable game than mines and CTD's, heavy lags and other weird errors...

[LarkinVB]

2. Perhaps 8 pop is a good minimum limit before teching ? Not sure.

3/5. My initial eldar BO (DoWpro) looks like this

	eldar_race =
	{
		{ "flag_capture" }, { "flag_capture" },
		{ "flag_capture" },
		{ "squad", "eldar_squad_bonesinger" },
		{ "squad", "eldar_guardian_squad" },
		{ "squad", "eldar_guardian_squad" },
		{ "squad", "eldar_guardian_squad" },
		{ "research", "eldar_fleet_of_foot_research" },
		{ "post_builder" },
		{ "building", "eldar_aspect_portal", true },
		{ "building", "eldar_warp_generator" },
		{ "squad", "eldar_squad_farseer", true },
	},

5. Was just an idea to improve CC. If turrets are in range (20) and enemy ranged squads have a minimum strength counting them shouldn't be too inaccurate.
Another idea is to avoid armed buildings at all with CC units during first 5 minutes.

Why is it impossible to count mines ? Aren't they handled like other buildings ?

Edited by LarkinVB, 23 February 2006 - 08:29 PM.

[ArkhanTheBlack]

2. Perhaps 8 pop is a good minimum limit before teching ? Not sure.

I don't use pop anymore since it's to inaccurate. I only use army strength/cost for comparison.
Normally the AI's build up a force of 800 in tier 1 and then tech to tier 2. But that's not enough. After 4 minutes they tech to tier 2 and stop all troop building if necessary, except their HQ is attacked. In WA this is a necessity, even if they've only one squad left. If you haven't tier 2 after ~6 minutes you're dead in WA. In DOWPro it's probably a bit different.

3/5. My initial eldar BO (DoWpro) looks like this

I'll move it directly after the first generator in the build program. This should be pretty much the same research time in 1.7 AI scheme.

5. Was just an idea to improve CC. If turrets are in range (20) and enemy ranged squads have a minimum strength counting them shouldn't be too inaccurate.
Another idea is to avoid armed buildings at all with CC units during first 5 minutes.

Normally the key is the commander. He normally has an extremly good armor type against turret fire and the other squads are quiet save if he takes the fire. But the attack strategy/plan isn't so clever, it compares only squad/army strengths and turrets/fortified posts. But coding that is very difficult and requires an insane amount of testing. Movements have to be coordinated so they fit attack/defend and retreat behaviour. It's really the hell of AI coding. Believe me, I know it.

Why is it impossible to count mines ? Aren't they handled like other buildings ?

Relic bug! It's a while, but I think the building count always returned 0.

Edited by ArkhanTheBlack, 23 February 2006 - 09:26 PM.

[ThetaOrion]

Ah, before I forget it... I removed mines only because I couldn't control how many are built from them. If they start once, they'll never stop. If you consider 7 AI's and each building 100 - 200 mines, you can get 1000 - 2000 mines on the map. Since I can't guaranty stability with such insane mine counts, I decided to remove them. Better no mines and a stable game than mines and CTD's, heavy lags and other weird errors...

Thudo said that he had a way to do the mines that addressed this complaint, by making the mines progressively more expensive as each faction tiers up, iirc. This solution of Thudo's sounded like the AI would do a few mines, suddenly find them more expensive and stop doing them, and then probably wouldn't do any mines at all, unless it is at Tier 4, fully capped, done building, and has resources piling up and not being used while the builders are standing around doing nothing in Tier 4, then it will save up and start buying and building really expensive Tier 4 mines. Making the mines really expensive in the later tiers should prevent their being 2000 of them, the AI will build the cheaper stuff first.

1.60 of the AI mod was handling mines well enough, I thought.

But, 1000 to 2000 mines would indeed be a CTD problem and possibly a path finding problem as well.

So, 1.60 of the AI Mod was producing CTD's due to mines on some people's systems?

--

Why is it impossible to count mines ? Aren't they handled like other buildings ?

Of course, counting mines and putting in a cap of 20 or 30 mines per player would be another way to go if it is possible to do so.

Edited by ThetaOrion, 24 February 2006 - 12:40 AM.

[Zenoth]

Phew ...

I've been quite busy lately at work (I still am, and will be for the coming week ... there's moments like that in life ya know), so I couldn't find the time to test Beta 11 properly.

The map was Doom Spiral. Difficulty level Hard, number of players 4. No -dev switch. Resources rate normal, and Resources sharing off.

It was me (SM), allied with Imperial Guards­, against Chaos & Orks.

I and my IG ally won.

Technically, I noticed nothing wrong. At least not in terms of movement behavior (such as back and forth issues as experienced in Mountain Trail at Critical Locations). There's a CL in Doom Spiral, however, as I mentioned, no gathering problems at all like in Mountain Trail.

I'll have to take a look at the replays posted in this thread by Theta, since I pretty much missed everything related to Beta 11 testing so far (sorry about that guys). I'll take the time now to read the whole thread's posts. And be sure not to repeat issues that were already explained/reported.

As for my test, well, I post the replay here. I'll come back at it later tonight, since right now I feel more like relaxing somewhere else than in front of my monitor.