61 replies to this topic

[Flenser]

I use the Avatar in one of two ways. Either I keep him at the base or I send him hit-and-running.

If he is at the base then he is (a) providing his cap increase advantage and (b) protecting the base from sneak attacks. If he is out hit-and-running then he should not be engaging the main enemy force (that is what my main force is doing) so he is unlikely to get hurt too badly. If he does get hurt I revert to plan (a).

His bonuses are too valuable to risk him getting killed. If he does then it takes a loo-oong time to build a new one, and if he got killed it is highly likely plenty of your other squads got killed as well and you will be unable to get your army back up to its previous strength until another Avatar pops out the oven.

Flenser

[thudo]

Yes but we're talking about the AI at this stage in the game. End-game for our AI is a joke now as its uber. Early-Mid game isn't. Its still, tricky. Everything is perky at the latter stages. However, I guess we could go both ways - keep Avatar at main base for protection or just have him stay around captured/fortified LPs then retreat to main base if at 25% health. I'm good either way. Usually, when I get ubers I'm raking in so much coin I couldn't care less what happens to them and the AI shouldn't either - its late game! If you allow the AI to tech that far then prepare for the consequences especially with SM/CSM.

[Quitch]

Hey, I only come here to help since it's obvious that certain members of this project, while being good coders, have absolutely no idea how DoW is actually played.

Make them build turrets around their generators, but it will be a complete waste of resources. The later the turret is built the bigger the waste.

The SM Landraider is useful and therefore the AI should be considering building it, but only in the situations where it is useful i.e. anti-infantry

Frills are something AIs should avoid. No one cares about "cool" AIs, they toy with them, then go download one that will give them a fight. AI scripting limits you enough without imposing handicaps on the AI because you want to see the "cool" units on the battlefield. There are many useless units (Bloodthirster, Avatar, vanilla Terminators etc.) and the AI simply should not build them!

I look forward to the next release improving on the last. It will need to, I saw the comments the last one got, many people were excited about it, downloaded it, fired it up... then it built a lot of turrets like a noob and got wiped from hard drives across the world. I tried two games with it before I couldn't take any more. The vanilla AI has its flaws, but its early aggressiveness is one of the good things about it, it doesn't try to turtle because turtling doesn't work outside of critical point turret spamming.

What's worse is that the turret strats made it in even though some within the team KNEW that this was not how DoW played, and the other members only needed to download a few top level replays to see this. Even now, with a release candidate tossing around, Rangers were part of the Eldar opening build... WTF?!!

I know some will take offence, but I really want to see this project succeed. It has some really good ideas, but I've seen other AIs with great code let down by really, REALLY shitty builds. The code is only as good as the units the AI uses, and if you pick shitty ones then it will suck.

The reply above shows a basic lack of understanding regarding this game. No one who understands the Avatar's use sends it into battle. No one who knows Chaos builds a Bloodthirster unless the game is already in the bag. No one who is good with Eldar uses Rangers in their opening build. The list goes on... and you should never have that many resources, not unless you've already won or it's quick start.

Edited by Quitch, 28 July 2005 - 09:53 PM.

[oozish]

Quitch, have you even played release 1.2? You know it's out right?

[LarkinVB]

He seems to mix things up but is right with some comments regarding how the game should be played. No turrets, no rangers.

What I don't like is his attitude. I could accept it if he has to pay for this.

[Quitch]

Call it my passion, its fustrating, especially as since this AI project exists there won't be another, so if this one f***s up, then that's it for the skirmish scene. You don't get more than one AI project per RTS game, not where a decent scripting engine is concerned.

Its fustrating to see so many obvious mistakes, as least judging from the board comments, being made when so much material is available that the build orders should be a piece of cake. In terms of that this game is pretty simple, especially in 1v1. After the first release of this AI almost every comment I saw mentioned the turrets, yet here we are, with a discussion on how the AI should build turrets mid-game, how it should protect generators with them. It just stinks of someone coming from a game like Warcraft III... even the language used when discussing this AI on the very first day "It should protect its expansions with turrets"... WTF is an exapansion in DoW? They don't exist! Turrets don't work in this game like they do in Warcraft III and you shouldn't try to shoe-horn them into that role. If turrets are still being used then it simply shows the lessons aren't being learnt properly from earlier releases and that is very fustrating.

Sorry if I come off as offensive, as I said I just want this project to succeed, but since modders don't work for the greater good they do it because its fun for them, don't expect to pass off that "its free" rubbish on me

The 1.2 topic wasn't up when I posted. I sure as hell will be giving it a try

EDIT: Okay, turns out I'm blind... damn, and I'd been watching for it too

Edited by Quitch, 28 July 2005 - 11:56 PM.

[thudo]

Quitch.. What the frick'n ever. Shheshh.. Yeaaa that attitude blowz royal donkey a*s.

Gawd damn you think yer f**ing almighty and yet we're coders/gamers who play the game constantly to ensure we ship a decent product AND YET you think perhaps Skirmish AI should be god-like? HOW MANY DAMN PC RTSes out THERE EVER HAD HUMAN-LIKE AI or was ever modded for?

Ziltch, buddy! Nil.

I love your logic - later the turret build the larger the waste. BUDDY -> it requires the AI to frick'n have 400req/400power to build one turret!!! And even then he's building units first, their priority. He's not losing games on turret building as I KEEP STATING he has THE COIN BY THEN so its being spent on units and turrets as a side-project.

And ubers.. gawd damn.. where is the LOGIC.. IF HE HAS THE COIN TO BUILD AN UBER HE HAS THE COIN COMING IN ! ! ! Not hard to understand basic economy + he has an extra barracks/machine pit to build off while the other is being tied down AND THATS if he has the coin too. Don't think we haven't covered most bases - we only play the damn game, oh, 2-5 hours a day in full test mode (at least I do kuzz its bloody fun).

Look.. its all about economy... if he has it late game THEN WHAT THE HELL IS THEIR EVEN A DEBATE FOR? DoW is about establishing early-mid game dominance. If you can't score the deciding below before late game then its a war of attrition scenario! So then at that point with stalemate one can introduce new elements. An uber is one of those.

Get me all hot and bothered about shiet like this.. its a true passion of mine and I'll debate you to hades and back about some of our decisions. If you don't stand for something, you'll fall for anything.

[thudo]

Quiltch - this project won't fail as its like no other ever seen in PC RTS land.. The AI code is quite unique and fully scalable/modular. We've had our differences and rattled each other and then Relic released v1.30 and almost sinks us. Then I figure something out 3 weeks ago, put us back on track 2months before WA, now we can get v1.3 out by end of August and then MAYBE v1.4 if WA is delayed a little.

No hard feelings guy.. its my passion too..

[Quitch]

The AI shouldn't turret... ever. If it has 400 req and power then it should keep it, or give it to an ally in a team game. Turrets late on are just fish in a barrel and take no time at all to kill. Hell, if you've got that much req and the turrets ever meet an enemy, that's a sign the req was needed to rebuild squads which died.

Again, you lack logic with the ubers. Having the money to buy something is not a reason to do so, it should be using that money for troops, research and upping the cap limits. Beyond that it should be saving the req for rebuilding its losses. There is never a reason to build the Bloodthister.

Don't worry about offending me (as if) I have the backbone to defend my point of view... I wouldn't post it if I didn't think it was the right one.

Edited by Quitch, 29 July 2005 - 12:48 AM.

[thudo]

Turrets don't cost 400/400, thats the requirement we set for it. Its steep. Enuff said.

Ubers serve a purpose - magnets! AI is always replenishing its economy so building and fighting with an uber is a moot point. With second barracks/vechicle buildings the AI is always producing. I highly doubt you'd be happy if you were in a war of attrition, an uber shows up, acts as a magnet while other forces eat you bare. Moot point.. Not worth arguing. AI never looses games when uber show as by that time their already established. Debate should be: how to consistently make the AI good so he gets better at starting up then moving to the higher-tiers.

[Quitch]

Turrets don't cost 400/400, thats the requirement we set for it. Its steep. Enuff said.

Ubers serve a purpose - magnets! AI is always replenishing its economy so building and fighting with an uber is a moot point. With second barracks/vechicle buildings the AI is always producing. I highly doubt you'd be happy if you were in a war of attrition, an uber shows up, acts as a magnet while other forces eat you bare. Moot point.. Not worth arguing. AI never looses games when uber show as by that time their already established. Debate should be: how to consistently make the AI good so he gets better at starting up then moving to the higher-tiers.

<{POST_SNAPBACK}>

400/400 is not steep and building turrets with those resources is a waste. It should be building a second machine pit, or ensuring every squad is maxed and its cap limits hit... it isn't doing that so the code is flawed.

LOL, magnets? You know what I fear more than a unit that acts as a magnet? That those resources are spent on a unit that will damage my army. Heard of cost-effective? Yep, ubers like the Bloodthirster are NOT cost-effective and SUCK as tanks because they cost you more to use them in that role than it costs me to take them down and each time you build the Bloodthirster you are in fact defeating yourself. You may fear it, but then that shows you simply need to play the game more, not that everyone else is wrong and the Bloodthirster is a really great unit.

The avatar is too useful to risk in combat.

Changes to the build order look good, I'm looking forward to it. I got the latest release and was nicely surprised by the variation in the AI behaviour from game to game.

I am going to stick my oar into the general slanging match that seems to be developing. Here goes....

1) It would be possible to code the ultimate uber AI, that did exactly the same thing every time and owned you every time. I don't think this is what the project is all about. I really like 'cool' AI, and no matter what you say Quitch, everyone has a little hiccup in their battle plans when a bloodthirster stomps your way. And it looks cool.

2) turrets in the right place buy time. I agree that you don't need 20, or even three, but a correctly placed turret can buy enough time to send something to actually deal with the problem. The turret isn't there to stop the attack like in (For example) C&C generals, but it can be used to delay the storming of the base. I'm guessing that was the thinknig behind the turret placement.

3) wartukks, rangers, and the other debated units are good fun to have in the game. (Sorry, I mentioned the word FUN) and add variety to the whole thing. Oh, and wartrukks could come into their own if they can be used as they are meant to be by the AI. - Instead of walking the big mek and slugas into the base, why not drive and arrive refreshed, ready to beat the hell out of a guardian/space marine/ raptor etc. That would be fun AND cool.

I'm not taking 'sides' or trying to increase the amount of arguing that goes on. Having read the thread it seems that Quitch and the AI team both want the same thing - A better AI. However, there are two different ideas as to what makes a better AI, Quitch wants an AI that is a damn good 'player' and the AI team want an AI that uses all of it's varying potential.  I want an AI that does different stuff and throws some surprises into the mix. If that means the AI turns up with trukkloads of shootaboyz and does a 'drive-by' on my bonesinger then so be it.

Now that would be fun AND cool AND pretty bloody annoying 

Keep up the good work guys,

<{POST_SNAPBACK}>

1. Replays look cool, the AI should challenge. If the aim of this project is for "cool" then all this dance code, all this team work code, is a waste of everyones time since the build order will destroy the AI and the rest will go to waste.

An AI project should aim to make the AI fight as best it can, otherwise there's nothing "intelligence" about it. If that means the same build order everytime, then that's just the way it will have to be and you should be blaming Relic for forcing that kind of stratergy on the game. There is still enough variety in this game that no one build order works for everything.

2. Turrets work early, and only early. They are outranged by most heavy weapons and vehicles and are very, very easily destroyed. There are only a few select times turrets should be built, and if they're your idea of defence then you're in for a shock. LPs are your turrets in this game. Not to mention that turrets won't stop me completing T&H or AC, and they don't count towards your Annihilate buildings so can be ignored completely.

I've only seen them used well twice, both times on Fallen City (about the only time they ever get used). Both times it was by the players near bridge and both times they'd been upgraded to missiles to act as a bit of a shock to Ork wartracks that were raiding while the SM army was on the other side of the map.

3. Oh FFS, can we shove this "fun" nonsense? What a waste of the coders time. "Fun" is when the AI completely kicks my ass, outplaying me, outmicroing me and making me sit back in my chair in awe of the way it played. If it's going to build shit then it's going to get beat and you won't see the "cool" ubers, because it will be dead by then. You won't see the cool tier 2, it'll be dead. You'll barely see the rangers do anything because they'll be running from my "not fun" build design to do what this game is about: completing the objectives.

The scripting language is what limits this generation of AIs, it doesn't need help from the team, they should be fighting to make it as good as possible and letting the difficulty levels take care of the rest. The second "fun" units make an appaearence, this AI becomes a novelty, not challenging enough to play, not as good as a replay to watch, just something that a couple of offline players can sit back and admire the animatons of... and you can do that with replays already.

Edited by Quitch, 29 July 2005 - 09:44 AM.

[LarkinVB]

If you can drop your teacher like attitude I may consider reading your text more carefully. Don't tell me the definition of fun or waste of coders time. Stick to your shoes and don't get annoying.

I'm working on a strong AI and I know how to play DoW properly.

[Quitch]

My attitude will not affect the accuracy of my comments and so frankly should be the least of everyones concerns. I came here to help this AI improve, not to pander to fragile egos. You may ignore my comments as you desire, but I will post my thoughts as long as I am able.

Edited by Quitch, 29 July 2005 - 10:28 AM.

[LarkinVB]

But you are of no help with your attitude. So troll off, bub and settle down at Relics forum. You are a guest here. Its our devs forum and you should behave. The accuracy of your comment is of no value as long as you don't find the proper style to spread it.

My ego is fine, thanks.

You may ignore my comments as you desire, but I will post my thoughts as long as I am able.

Deal.

Edited by LarkinVB, 29 July 2005 - 10:47 AM.

[Flenser]

Quitch maybe expresses himself in a way others on this forum do not, but I do think he is being negative, disrespectful or insulting, and I think he has some valuable points to get across, and we ought to listen.

His points about a fun AI vs. a good AI are very valid, and we are possibly guilty as charged of - on occasion - making an AI do stuff we think is cool, rather than making it the best it can be. Hell - I don't KNOW that Skirmish plans are helpful, I just did them cause they were fun. Now, it is our MOD and we can make it cool if we damn well like, hell, we can make it ice cold, but Quitch is at worst guilty of making sure we are all aware of the distinction between cool and good, and that is worth hearing, and I have no objections to his tone. He has even apologised for giving any offense - trolls don't do that.

It is all well having an objectives list, but there are higher objectives in this MOD and we need to remember them, recognise them and make sure we are happy about them.

And Larkin, Quitch is saying nothing about turrets that you have not said a hundred times already. I thought you would be recognising a like-minded player rather than asking him to take it elsewhere.

Flenser

[Quitch]

After v1.0 Larkin did welcome me, with open arms, as I banged into Thuds skull what he'd tried to do before. He doesn't like a direct approach, some people don't, but I don't beat around the bush. If something is total arse then that's what it is, and no amount of delicate phrasing will change that.

Edited by Quitch, 29 July 2005 - 12:00 PM.

[oozish]

Quitch.. What the frick'n ever. Shheshh.. Yeaaa that attitude blowz royal donkey a*s.

Gawd damn you think yer f**ing almighty and yet we're coders/gamers who play the game constantly to ensure we ship a decent product AND YET you think perhaps Skirmish AI should be god-like? HOW MANY DAMN PC RTSes out THERE EVER HAD HUMAN-LIKE AI or was ever modded for?

Ziltch, buddy! Nil.

I love your logic - later the turret build the larger the waste. BUDDY -> it requires the AI to frick'n have 400req/400power to build one turret!!! And even then he's building units first, their priority. He's not losing games on turret building as I KEEP STATING he has THE COIN BY THEN so its being spent on units and turrets as a side-project.

And ubers.. gawd damn.. where is the LOGIC.. IF HE HAS THE COIN TO BUILD AN UBER HE HAS THE COIN COMING IN ! ! ! Not hard to understand basic economy + he has an extra barracks/machine pit to build off while the other is being tied down AND THATS if he has the coin too. Don't think we haven't covered most bases - we only play the damn game, oh, 2-5 hours a day in full test mode (at least I do kuzz its bloody fun).

Look.. its all about economy... if he has it late game THEN WHAT THE HELL IS THEIR EVEN A DEBATE FOR? DoW is about establishing early-mid game dominance. If you can't score the deciding below before late game then its a war of attrition scenario! So then at that point with stalemate one can introduce new elements. An uber is one of those.

Get me all hot and bothered about shiet like this.. its a true passion of mine and I'll debate you to hades and back about some of our decisions. If you don't stand for something, you'll fall for anything.

<{POST_SNAPBACK}>

ya, once the AI get's ahead of you, the resources are not an issue, you'll have none, he'll have it all...so why not go down at the hands of an avatar? That would be cool. I've not yet, well maybe one time, come back post mid game after the AI get's ahead. Usually there is simply no stopping it.

That's why it's a non issue...the end of the game is neigh, there is no 'coming back' by the time the AI is deciding these things. And the Ai will have MORE RESOURCES THAN IT CAN SPEND.

Still, one point to be made, figuring out late game strat for the AI is probably time misspent since the focus should be on early game decisions still and fixing the mid game problems such as vehicle productions etc.

Quitch, you're obviously quite smart and well versed in DOW, however, what you seem blind to is that your attitude and manner is a big turn off. You must respect those who do this in their free time and (for free); and your 'tone of voice' in text is not respectful. Plus, you've obviously pissed off thud, and that's hard to do, so tread lightly if you're still around in a few days maybe you can try a different approach, I think I 'heard' it in your latest posts and you seem to be coming around.

Cheers.

Edited by oozish, 29 July 2005 - 01:39 PM.

[Quitch]

I've done AI for other RTS games in my free time, for free, and I don't expect people to kowtow to me for that, be it on their forums or mine, and I don't kowtow to others either.

My manner is one of strong feeling and a direct approach. Some don't take kindly to it, but then that will be their loss.

DoW is a game of comebacks and being ahead in resources does not indicate victory. When I first tried Ork against a Harder Chaos in 1.2 I lost virtually the whole map (Battle Marshes) yet came back to win the game through superior micro and tactics. If the AI had been building defensive installation during a time when I was the one on the back foot, it would have handed me the initiative and I would have won three times as fast.

Time on the late game is never time misspent. I've seen countlesss replays of good Eldar players losing if the game drags out, not because they were outplayed, but simply because once the game hits a certain stage they're in unknown territory, they don't know how to play the late game and they lose.

The AI will never have more than it can spend unless it's doing something wrong. Troop losses and SP decay will see to that.

Thud further clarified his point in his next post, in other words I will push my view point hard and he won't give ground with ease. That's the way it should be.

Edited by Quitch, 29 July 2005 - 01:48 PM.

[thudo]

Yes but you must be realistic about our project:

1) Skirmish AI for any PC RTS does not see "the big picture" and cannot dynamically alter its personality to suit a new attack style. Oh it can see the entire map but it simply cannot switch to something risky. Scripting IF and THENs can only go sofar until we get 10Ghz Muti-Code CPUs or an AI processor.

2) You actually seem to be saying many positive things about the AI but please remember to just keep things real for us. I know you don't know what is realisitic for us to accomplish and trust me: I'd die to include some of your suggestions but no SKirmish AI project EVER has done true learning dynamic AI. Things must be kept "down to earth" for us as it can easily spiral outta control.