Jump to content


Photo

Final beta 8


110 replies to this topic

#101 ThetaOrion

ThetaOrion

    title available

  • Members
  • 676 posts

Posted 14 February 2006 - 08:14 AM

Well, I played Kasyr Lutien with and without -dev, 3vs3, using the same build order, HARD Beta 8.

I as Chaos 1, Orks 2, Chaos 3 AGAINST SM 4, Eldar 5, Imperial Guard 6.

This time around, I really didn't notice much difference, except I lasted a lot longer with -dev. In both scenarios, with and without -dev, the Eldar and IG converge on the human player early in the game. The IG have tanks, Sentinels and Hellhounds, and I have yet to build a Machine Shop. And of course, they outnumber you 4 to one on troops. After my base is destroyed, I build again elsewhere.

At first glance, this particular scenario is more in line with what Excedrin reports, aka no real difference with or without -dev. BUT, it's a subtle thing that only becomes apparent over a long period of time, on the larger maps for the longer haul, the difference is still very noticeable.

Where it differs is that with -dev, my AI allies hold and eventually take down the Space Marines. Without -dev, the Space Marines and none of the other AI enemy bases are ever really touched (even though I tried without -dev twice) and my AI allies are losing big when I quit. You and your allies last a heck of a lot longer with -dev, which means that it is a lot more balanced in the long run with -dev. With -dev, I still had AI allies, and the Space Marine enemies were gone. It was more equal all the way.

But, in both scenarios, my production is nil by the time I have built my second base, and I quit the game. It's starts getting boring for me. And, with -dev, I almost get the feeling that the game could go on for another hour or two. Without -dev it's over in less than 30 minutes, and we lose big. Without -dev, it feels more like a HARDER setting, even though it's supposed to be only HARD.

--
||
--

SECOND MESSAGE:

Test same map with same setup AI vs AI only with and without -dev.


How do you do an AI vs AI without -dev? :D

I thought that Thudo said that it was mandatory to use -dev in order to make it a pure AI vs AI match.

Edited by ThetaOrion, 14 February 2006 - 08:15 AM.


#102 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 14 February 2006 - 08:16 AM

Test on big maps with players involved and lots of random factors are of no real use for the '-dev' discussion.

You do know that a battle can have different results because a single squad fight at gamestart had a different outcome by random ?

Remove the most random factor of all first : human player. Only use AI vs AI.
Remove the lua code randomness by using fixed math.randomseed()
Start testing 1 vs 1 with/without -dev, record the game and make notes.
Continue with 2 vs 2 and up.

Something like "my ally lasted longer with -dev in my 4 vs 4" is irrelevant as the random factor is much too big.

Edited by LarkinVB, 14 February 2006 - 08:25 AM.


#103 ThetaOrion

ThetaOrion

    title available

  • Members
  • 676 posts

Posted 15 February 2006 - 01:56 AM

LarkinVB, I don't know how to do AI vs AI without -dev. I can do AI vs AI with -dev, using those autoexec.lua commands that Thudo told us to remove from our autoexec.lua.

But, if with and without -dev experiments have no further value, there's no sense doing them. The world will be playing 1.70 without -dev, so that's what I'm going back to.

And now that I think about it, I didn't sign up to do AI vs AI testing. Thudo and Arkhan can do those. They don't need me.

I have indeed been rethinking my involvement with this project. In the past few days, Thudo has provided numerous and compelling reasons why this project would be much better and happier without me and without my input.

I'm just trying to rethink my future involvement if any. The revelation about -dev and Thudo's input has convinced me that my input over the past couple of months has done more harm than good. I believe it.

Thudo: Sick and Bloody tired of this move toward unknown maps. WE NEED A FOUNDATION to ensure the Official Maps work well with the AI and then.. MAYBE we will springboard elsewhere. However, Official Maps MUST be tested first to ensure consistency and a solid base. Thanks all!


Most of the wierd stuff regarding the Beta AI (with and without -dev) only shows up in the largest maps and/or the longest games -- stuff like endless fields of unfinished Eldar turrets. Plus, I do believe that the whole world will be playing 1.70 on unknown maps, including some of these larger maps, Relic official or user-made. The only thing that makes this game new and interesting for me personally is all the new maps, so I'm not going to stop playing them. They were my primary interest before I got involved with betatesting the AI Mod.

A future plan, where Theta is concerned. I was thinking that if I'm permitted to continue betatesting the AI Mod in the future, if I test some future Beta without -dev on the five approved maps (Mountain Trail, Battle Marsh, Blood River, Fallen City, Valley of Khorne), after watching the replay, if I see something goofy, I'll just upload the replay without comment and let them find the problem themselves. Replays with a user in the mix is the only thing they can't get from AI vs AI, and with a human in the mix, the game does at times act differently as you LarkinVB just suggested. The repeated request for me to stop commenting is the message that comes through the loudest. I need to try to comply.

On the larger maps 6 and 8 player maps and the unapproved user maps, I just won't mention the bugs that I see. I'll let the users of 1.70 make their complaints then. I thought about taking pictures of the wierd stuff like the endless fields of started but unfinished Eldar turrets, zipping it up and uploading it without comment, but I don't know if that would be useful or not. No sense for me to put in time doing stuff that has no value to anybody or that actually makes people mad.

Of course, if Thudo wants me to completely go away, I can do that as well. A couple of months ago, Thudo told me to come in and have fun. Now that it is no longer fun and feels like work, if he wants me gone, I can do that too. It's up to him. I'm here at his invitation, and I can go away at his invitation, too. I don't want to be where I'm not wanted and not needed. There's no fun in that.

#104 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 15 February 2006 - 04:15 AM

Ok, I've been busy lately and couldn't make reports.

Now, on to the -dev Vs. without -dev subject ...

Today so far I had the time to test some more (It wasn't A.I Vs A.I tests though).

I played in 1 Vs 1's, 2 Vs 2's, 3 Vs 3's also 4 Vs 4's. Some of them at the Standard/Normal difficulty setting, and the majority at Hard.

The more I test, and the less I see noticeable differences between my matches without the -dev switch and with it.

Indeed, I reported during my first tests with the switch that I noticed quite better A.I behavior. That the A.I committed less mistakes, such as, for example, a "back and forth" movement between specific locations in a given map, in a given context (such a situation is actually seen in one of my reports for Beta 8, the one without the switch, where I noticed a problem regarding Imperial Guard's A.I behavior in Mountain Trail, around a specific Critical Location).

Well, in my tests with the switch, the problem occured. Of course again in Mountain Trail. Very same technical issue, which tells me that, at least regarding that very specific A.I "error", the -dev switch did not have any impact or influence whatsoever.

With that observation alone, I can already start to think that all the changes/differences noticed in games with the -dev switch, compared to those without it, were probably nothing but coincidence.

Right now, at the very moment as I type this, that's my current conclusion on the subject.

That's where comes this question for Larkin ...

What are the exact steps I need to do for A.I Vs A.I testing ?

I want to do the tests and report every single thing I may notice.

But only changing the math.randomseed(1) line is surely not going to suddenly make all my matches A.I Vs A.I only, is it ? I tried it anyway (the change in the code line as you said), and upon entering the Skirmish menu as usual I was still part of the game. I was still the player #1.

What do I need to do technically, for A.I Vs A.I ? Sorry but I need to ask since I never did that before. And we all need to start somewhere.

Once I know how, then I'll do my best to report what must be. So we can move on, and perhaps fix a few last things (like the movement problem I noticed) and release it to the public.

Edited by Zenoth, 15 February 2006 - 04:19 AM.


#105 ThetaOrion

ThetaOrion

    title available

  • Members
  • 676 posts

Posted 15 February 2006 - 05:27 AM

For me, the difference with -dev and without -dev only shows up in team play on the larger maps, and the really wierd Beta 8 stuff only shows up in the very long games, aka the unapproved 6 player and 8 player maps. There's only one approved team play map (Mountain Trails) where the difference between -dev and without can be demonstrated and accepted by the leaders and oldtimers. The phenomenon is currently discounted or dismissed as coincidence, if you see it on any other map.

However for me, without -dev, it's like it is all locked into the same build program. I get the same losing game over and over again -- I'm at HARD setting, the setting I selected, the AI enemies are at HARDER production settings with a bonus, and the AI allies are STANDARD setting in terms of production. This is on the approved Mountain Trails Map, but only with the affected factions, aka Space Marines and Orks and IG. SM enemy goes uber and Ork AI allies go wimpy. On the same map, without -dev, I beat the crap out of everybody if I play as 2 Eldar against 2 Chaos. Then everything seems normal to me. So, is the problem with -dev, or is the problem with the SM and the Orks and the IG? Who knows? But for me, the gameplay changes noticeablely with and without -dev on Mountain Trails with Chaos 1, Orks 2, AGAINST IG 3 and SM 4.

However for me, with -dev, the long games on the larger maps play out more equitably, and it looks like I'm getting a different set of build programs than the ones I'm repeatedly being locked into without -dev.

It's subtle stuff, and most the time it only shows up over the long haul, after 40 minutes of clock time. If I want to feel or see the phenomenon, it requires a balanced map, balanced equally sized team play, and the larger the map the better. And for me, the SM, Orks, and IG have to be involved. Again, lots of ifs there, which makes it all very easy to discount and dismiss as simply coincidence.

EDIT: The AI vs AI Cage Matches are a better test of the AI's readiness imho. They are much more accepted, more incontrovertible, less subjective, and remove the human emotion from the equation. Humans can be discounted and dismissed. But, the masters need to remember that 99 percent of the people who play 1.70 after it is released will be playing without -dev and will be playing with a human in the mix.

Edited by ThetaOrion, 15 February 2006 - 05:54 AM.


#106 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 15 February 2006 - 07:03 AM

Sorry, I completly forgot that AI vs AI isn't possible no more on a single computer without -dev. Of course you can have AI vs AI without -dev with two computers.

@Zenoth :

You get AI vs AI with -dev while adding this line to autoexec.lua in the WA root folder. If the file doesn't exist you have to create it.

 run = Cpu_ControlLocalPlayer()

With these additional lines you can set gameplay spped with shift F1 to F5

bind( "shift+F1", "vslow()" )
bind( "shift+F2", "slow()" )
bind( "shift+F3", "standard()" )
bind( "shift+F4", "fast()" )
bind( "shift+F5", "vfast()" )


function vslow()
	setsimrate( 1 );
end

function slow()
	setsimrate( 6 );
end

function standard()
	setsimrate( 8 );
end

function fast()
	setsimrate( 15 );
end

function vfast()
	setsimrate( 30 );
end

Edited by LarkinVB, 15 February 2006 - 09:36 AM.


#107 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 15 February 2006 - 07:34 AM

Sorry, I completly forgot that AI vs AI isn't possible no more on a single computer without -dev. Of course you can have AI vs AI without -dev with two computers.

@Zenoth :

You get AI vs AI with -dev while adding this line to autoexec.lua in the WA root folder. If the file doesn't exist you have to create it.

 run = Cpu_ControlLocalPlayer()

With these additional lines you can set gameplay spped with shift F1 to F5

[/code]
bind( "shift+F1", "vslow()" )
bind( "shift+F2", "slow()" )
bind( "shift+F3", "standard()" )
bind( "shift+F4", "fast()" )
bind( "shift+F5", "vfast()" )


function vslow()
setsimrate( 1 );
end

function slow()
setsimrate( 6 );
end

function standard()
setsimrate( 8 );
end

function fast()
setsimrate( 15 );
end

function vfast()
setsimrate( 30 );
end
[/code]


Ok thanks Larkin.

I'll try it on Beta 8 before moving to Beta 9.

#108 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 15 February 2006 - 08:33 AM

Just tried A.I Vs A.I, with the -dev switch. It was a 1 Vs 1, Hard, in Blood River (is that map official ? I don't remember if I had it after installation or from one of the many maps I've downloaded). It was Imperial Guards against Chaos. Resources Rate at Standard.

I could not see noticeable difference without me in the game, than with only two A.I's, at least not in that game anyway. Both IG and Chaos' behavior was the same I ususally experience when I play.

I can't even report anything, since nothing "wrong" happened. Both factions proceeded with usual build orders, with some randomization thrown in. They both produced vehicles at about the same time, evolved their technologies, upgraded their groups' weapons, attached lesser troops to higher characters (heroes and such). No "back and forth" movement. Good response time from away troops when base was under attack. Capturing CL's and LP's without problems. Re-capturing them when possible upon loosing them. Repairing structures.

And so on ...

Really, nothing to report.

Of course that's only from a single game.

But I won't be able to compare A.I Vs A.I with and without the -dev, since I do not have two computers, and I can only test A.I Vs A.I with the -dev switch, so, sorry about that. I would have liked to, but I just can't.

And, since Beta 9 did not include any game-play/A.I changes, I installed it, and the test I describe above was with Beta 9. I also saved the replay of that game, but as I said ... there's nothing to see nor report. Everything went smooth, until the end, when the Chaos won. But the IG made a good fight, and even had some superiority during the first moments.

I could, however, test more A.I Vs A.I games, with the switch, if you coders believe it'd be necessary. I wouldn't mind. I'm here to test. I'll do it if you want to. Such as 2 Vs 2's, 3 Vs 3's, etc. But I personally doubt we'd see noticeable differences.

I still think, right now, that what I myself considered as "much better behavior with the -dev switch" might well have been coincidence.

Edited by Zenoth, 15 February 2006 - 08:36 AM.


#109 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 16 February 2006 - 07:00 AM

Ok ... I have tested some more with A.I Vs A.I.

For those tests (well, most of them, which means about 4 of them, out of six I think) I tested the Eldar, and tried to catch what ThetaOrion reported, which was fields of un-completed turrets.

Well, so far, I have yet to see the issue in question. All the games were tested in Mountain Trail and Blood River (2 Vs 2's and 1 Vs 1's). The games were always one or two Eldar(s) fighting against one or two Chaos/Orks. At the Hard difficulty setting, as I usually test in (I have done some Easy and Normal difficulty tests, but only a few).

Some games were won by the Chaos, and one of them was almost won by the Orks. Some were won by the Eldars. It was balanced to the extent that both factions (Eldar/Chaos) could win, and not always one. That, I liked. There's even a game where the Orks almost did it (as mentioned) in Blood River, but the lack of numerous Killa Kan doomed them. For one Killa Kan, the Eldar produced two Wraithlords, and since their troops are mostly melee (the Orks), they couldn't really hold their ground against Eldar vehicles.

As for the games with mixed factions (IG/Chaos/SM/Orks), usually the SM/IG combo wins, and of course the Space Marines mostly do the job. The Imperial Guards, however, are far from being weak.

But, again, I'd like to come back on the issue I reported some time ago during Beta 8, where the IG had a back and forth behavior problem at one of the Critical Locations in Mountain Trail.

That issue, I have yet to see it again. I saw it maybe in three games out of then (approximately). It's not happening everytime. But so far I only saw the problem occuring in Mountain Trail, no other maps at the moment. Why I want to come back on the subject ? Well ... because simply I fear that it might happen again. So, yes of course I'll keep testing with the IG mostly in Mountain Trail to try and catch the bug again.

Also, I have a question for Larkin/Arkhan ...

Does the math.randomseed line, when changed to "(1)", has any actual, proven impact on the A.I behavior ? Or does it have potential to ? As in ... say ... the build order changes ? Or they tech up slower ?

Because right now, I still do my A.I Vs A.I tests with the change (changed to (1)), and I am wondering if I should go back to the default math.randomseed state (I saved a copy of the file before I made the change just in case).

Edited by Zenoth, 16 February 2006 - 07:01 AM.


#110 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 16 February 2006 - 08:08 AM

Does the math.randomseed line, when changed to "(1)", has any actual, proven impact on the A.I behavior ? Or does it have potential to ? As in ... say ... the build order changes ? Or they tech up slower ?


A fixed math.randomseed has the impact that all calls to a random number with math.random will return the same result. You will always get the same build program for example. This is only usefull if you want to test two battles under same conditions.

#111 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 16 February 2006 - 08:16 AM

All right then.

To sum it up, other than the Imperial Guard's back and forth issue in Mountain Trail, I on my side of the tests I made so far have nothing else to report in terms of malfunction for Beta 8/9.

I'll keep testing until perhaps another build is released before final.

And of course if I catch up any other issue(s) I'll report.



Reply to this topic



  


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users