44 replies to this topic

[Quitch]

Heard of AMAI? Its a similar project for Warcraft III, complete with personalities, adapting stratergies and so on.

With their latest beta they released a version that allowed you to pit vanilla AI against AMAI. I found this very useful in two ways:

1. The team functions of AMAI are underdeveloped, and in many ways its new features could work against it in team games. It was good to see how it fared against the old AI as that acted as a baseline for improvements (i.e. you knew how hard/easy it was to beat, therefore you could judge AMAI against this rather than in the vacuum that human vs. AMAI games created).

2. It proved useful in higlighting which of the fancy features (like distraction attacks), which in theory should be useful, were in fact harming the AIs ability to perform.

While this function isn't as useful as it could have been thanks to the crap that is the 1.3 AI, I was wondering if it was even possible? If so I hope you'd release something that allowed people like me to do this. Assuming Relic fix their AI in the next patch (and I assume you've reported the nature of the problem to them, since I doubt many outside this forum are fully aware, simply putting it down to inferior AI routines in 1.3) such a mode would only become more helpful.

[oozish]

With it's predictable 'staple' early rush, I would bet the vanilla AI would win, but it still isn't any fun to play against. Would be fun to see though.

[Quitch]

With it's predictable 'staple' early rush, I would bet the vanilla AI would win, but it still isn't any fun to play against.  Would be fun to see though.

<{POST_SNAPBACK}>

An easy to beat AI is not my idea of fun. If unwritten rules are required to generate a challenge then the AI has failed. If the vanilla AI can win with its first rush then that is a flaw and should be fixed.

However, I am inquiring about such a tool so that such questions are not hypothetical.

[Flenser]

Simple answer - yes. You could do it.

The only way I can think of right now would be the brute force method of simply including all the original AI files and appending something suitable to their names - eg. cpumanager2.ai. Function InitializeAI() in default.ai would then choose one or the other sets of files to use.

I suspect there would be a LOT of file and/or constructor class renaming (both of file names themselves but more importantly the names within files themselves).

Flenser

Edited by Flenser, 29 July 2005 - 02:41 PM.

[thudo]

The AI under our project has more at stake then the vanilla one (which is plain dumb and "rushy" ). Sure we can make it rush you if you like (I've seen that countless times with ours) but our AI is NOT EASY WHATSOEVER. Vanilla AI is not easy either but ours is smarter all the way through.

I got a helleva lot of new Larkin code to implement and test - tomorrow I'll be sending it to our betatesters for appraisal. The Eldar buildorder has been tweaked (no Rangers) + 10-15 new changes to excite the senses. Also waiting on Flenser for some of his brilliant Eldar Hit&Run tactic code + T&H/CA logic.

We're making progress.

[Quitch]

I played a few games today, 1v1 and 2v2 on Harder. Beat it, but then that's not surprising because I'm just so great . I find it harder to draw conclusions from my games than from other games but I'm going to go over the replays to see what it did right and what it did wrong.

It feels easier, but then I want to find why that is. Are its BOs simply suiting my style? Then again, the Relic AI was easy, it just took longer because I broke its squads with flamers then teched for the win.

The Eldar definately need help. Rangers need dropping and Banshees could be safely dropped too... it was noted that they destroy LPs faster than any other Eldar infantry bar WS. Oh, and drop the Seer Council too... now we're talking streamlined

Eldar are all about Guardians FoFing all over the shop, except in mirror games where it's a WS fest. Can't test that though, I'm virtually pure SM through and through.

I don't want to see it do an all or nothing rush, but I do want it to be able to survive one itself. Hopefully the dance code, combined with some smarts to make it dance its own LP2s, should ensure it can never be rushed.

Is what I asked for possible BTW? Can you release a MOD AI Vs. Skirmish AI version. AMAI did it by making the first half of AIs AMAI AIs and the second half Blizzard AIs. As balancing in a vacuum is asking for trouble, so is testing. The best way to check any change is to test your release against what came before. If 1.0 could beat vanilla and 1.1 could beat 1.0 etc. you know you're making progress. One of the easiest things to do is harm your AIs ability to fight and not even realise you're doing it.

Edited by Quitch, 29 July 2005 - 03:04 PM.

[Quitch]

What is "DoW v1.20-v1.30 Coexistence REG Trick.zip"? I'm sure I read about this once...

[thudo]

It allows for both v1.20 and v1.30 of the game to run on the same system. I have both versions running but now just use v1.30 since our AI works near flawlessly on it. When WA comes out its byebye v1.20.

[Quitch]

What about my original question? Is it possible to pitch the vanilla AI against the mod AI, or for that matter any two AIs against each other (even using the same race for both)?

[LarkinVB]

It is possible ... with lots of coding work. I brought up the idea some month ago but there was no interest.

[Quitch]

Allow me to register my interest, just in case it comes up again. Testing one version against another is your best defence against regression.

[thudo]

Hmmm.. I'd REALLY like to know how you can have two versions of the AI running at the same time? I gather it could work by pointing to different .AI files then selecting the player to emulate that. However, hands down our AI would own the original especially since ours gets upgrades and uses tactics far better than the shiet vanilla one which is just mindless trite.

[LarkinVB]

I guess you 'just' have to use 8 different directories named by player_id and load scripts dynamically from these at gamestart.

If I get bored I will test.

[thudo]

I can see what Quitch would want such a thing: it would allow you to see one player play under what Relic intended vs the evolution that we've allowed to happen. However, try our STANDARD AI -> does it play like a vanilla DoW AI? It should to a greater degree? Yes it has some advanced logic running and constantly evaluating in the background but it doesn't have as many research items and other "costs" that come with higher skills. Look at the research area alone: it only researches req/power upgrades hence the reason it can field lots more units and why 9/10 times you'll see an uber-unit on STANDARD since cash is flowing in so fast as the Standard scripts preform less costly actions.

Saying that, Standard is just, well, standard. Having research and upgrades is key to survival many times and Larkin has now allowed for such flexible research to commence.

Still.. if time permits then yes we could see how both compare. Would be interesting to watch who comes on-top.

[Quitch]

Hmmm.. I'd REALLY like to know how you can have two versions of the AI running at the same time? I gather it could work by pointing to different .AI files then selecting the player to emulate that. However, hands down our AI would own the original especially since ours gets upgrades and uses tactics far better than the shiet vanilla one which is just mindless trite.

<{POST_SNAPBACK}>

Mmmm, you think? That's what I'd like to see. Your AI is better in many ways, but its been said that the vanilla AI is more aggressive early on and I'd like to see how this mod (jesus guys, give me an easy name to use) handles that. It's the small things that are of interest. It may win, but that doesn't mean it does everything better, just more... and without a test saying it will win is just an educated guess.

More importantly, I'd like to see it in a team game against the vanilla AI. AMAI was a really good AI, but then they put it in a team game against Blizzard and found, for all their fancy features, they'd completed failed to take into account the differences between good 1v1 play and good team play. Everyone always does.

Regression testing is the most practical use of this sort of feature though. You add cool feature X, but without testing against a controlled baseline (which humans aren't, especially as beta testers tend to be fans of the mod) it is often hard to guage whether the work is for the better or worse.

[thudo]

If we were paid, like Larkin mentioned, we'd put a little more "gusto" into it. I, however, beg to differ about our early game now that Larkin released his new code. Try rushing an Eldar player now when he almos immediately gets DR/WSes out? Not gonna happen with the vanilla AI.

The only reasons, ONLY, the vanilla AI is good early is kuz:

a) it doesn't have the burden of costs for items especially research
b) its dumb so it just goes for massing by production without strategy/tactics

With our tweaks especially most recent, I'd give our AI the benefit of the doubt. Bias of course but who wouldn't when its "our own baby"?

[Quitch]

Which is exactly why developers make for cacky testers

[Danimator]

I've been away for a few days and have been trying to catch up with all the posts and this is quite a busy forum for the number of members......still not finished yet.

This thread has caught my eye though and just thought I would offer a suggestion or two.

If you want to seee how the skirmish ai does against the vanilla ai with out all the coding involved there is another way but it does have one main drawback....you can't play against the same team i.e. Orks vs Orks or Eldar vs Eldar.

However by playing a series of round robin tournaments you should get a very good idea of the results especially if the play the same matches with vanilla ai vs vanilla ai and compare all the result.

My suggestion is simply to strip out 2 of the races from the skirmish ai and then play the tournament. You wouldn't even have to strip them out just comment them out and change back when finished (or keep as a testing tool for the future).

As Thud says in another thread Chaos and Space Marines play a very similar game (compared to Eldar, Orks anyway) so take out one of these - Chaos and then either Orks or Eldar, lets say Orks.

You can now play skirmish ai Space marines and Eldar vs vanilla ai Chaos and Orks.

Another version could then be made to do it all the other way round.

Pretty simple to set up and interesting to see how for example Skirmish ai Eldar do against vanilla Orks compared to vanilla Eldar vs vanilla orks et. etc. etc.

I'm starting to ramble so I'll leave it with you for comments.

[LarkinVB]

Pretty simple to set up and interesting to see how for example Skirmish ai Eldar do against vanilla Orks compared to vanilla Eldar vs vanilla orks et. etc. etc.

Pretty simple ? What about the numerous generic routines that are in the advanced AI code which are NOT race related. You have to extract each and every line of code and add race checks. This is far more work than loading different sets of files from start.