Matrix Games Forums

I was just wondering if anyone is actually using these and posting them anywhere? The AI has changed so much that they felt like a good way to udpate some scenarios without having to learn the editor.

And just where does one find said 'AI files'. Ralph?

I guess I should have mentioned that other than the cryptic mention in the what's new...

The basic issue is that the AI in COW was broken, so a lot of designers didn't bother spending a lot of time on the AI. It's also a moving target and changes some every release.

It is meant to allow us mere mortals to 'fix' the AI of older scenarios, I added in a feature that would save off the AI objective tracks and reload them. It's very simple to use, and a lot less intimidating than the editor.

As a designer, this could be VERY helpful since you can change the objective tracks on the fly while playtesting a scenario, instead of having to restart each time. This should make for better overall AI.

I also added in the 'Elmer Assist' feature, and this could be important for that. If you wanted to play both sides of a scenario, you could play one side and save the objective tracks as what you want to have, then switch sides and play against the tracks you just set down to try different strategies.

One example of where I fixed this is "Red Thunder 1988". One of the bridging units at the bottom of the map had objective tracks set too close together, that meant that it never actually bridged the river, and the attack was stalled because of that. I was able to fix that scenario's AI without touching the scenario by going into it and changing that formation's path.

Ralph


From the what's new.
37. X hotkey saves a “scenario name.AI” file. This file contains the AI objective paths, perhaps edited using the Elmer Assist option (see below). If saved to any folder in the graphical hierarchy for the scenario, it will be reloaded whenever the scenario is run or loaded into the editor (be careful about this, designers). It also contains three settings for PO attack-odds limits as follows:

Threshold1: This is the odds (x10) required for Elmer to prosecute an attack with at least 7 rounds completed or if the Max Rounds per Battle is set. The default is 10 (1:1)
Threshold2: This is the odds (x10) required for Elmer to prosecute an attack with at least 5 rounds completed. The default is 20 (2:1)
Threshold3: This is the odds (x10) required for Elmer to prosecute an attack with at least 0 rounds completed. The default is 30 (3:1)

There is also a separate set of thresholds for the Berserk Strategic Bias. These settings would allow players or designers to fine-tune the PO in their scenario a bit more. But the file has to be edited in an XML editor (like XmlPad). See the “Red Thunder 1988” folder for an example of a .AI file.

ORIGINAL: ralphtrick

I guess I should have mentioned that other than the cryptic mention in the what's new...

The basic issue is that the AI in COW was broken, so a lot of designers didn't bother spending a lot of time on the AI. It's also a moving target and changes some every release.
...

Uhhh???

So, what you're saying is that the game was broken if you intended to play vs the AI?...

I haven't used the .ai files, but I'm weird - I like working in the editor and don't even know what xml is (extended metacarpal linguistics?).

... the AI. It's also a moving target and changes some every release.

Change is good. Is anyone actually working on COW Elmer?

... you can change the objective tracks on the fly while playtesting a scenario ...

I just wet my pants. [:'(]

Threshold1 ... Threshold2 ... Threshold3 ...

and ...

... thresholds for the Berserk Strategic Bias.

This stuff is over my head, but if it has anything to do with getting Elmer to act the same during a scenario as he does in the final 20% of the scenario, then I'm for it.

Just to see if anyone else agrees, I'd like to state that Elmer seems to know he has been surrounded/isolated and moves to meet the threat, but then he doesn't follow up by attacking to break out. In a case like this, Elmer's units are toast for the human player on the following turn, so the human has an advantage. Maybe Elmer should not only meet the threat, but should also go berserk in attacking to break the isolation.

Thanks for posting this stuff Ralph. More is welcome, I'll do my best to grasp it [:)]

ORIGINAL: jmlima

ORIGINAL: ralphtrick
I guess I should have mentioned that other than the cryptic mention in the what's new...

The basic issue is that the AI in COW was broken, so a lot of designers didn't bother spending a lot of time on the AI. It's also a moving target and changes some every release.
...

Uhhh???

So, what you're saying is that the game was broken if you intended to play vs the AI?...

Try playing COW again. Elmer is a limp dishrag. If you are next to an objective he ignores you. If take an objective behind him, he goes running back. To get around these issues, the designers tried plastering the map with objectives and doing other things that didn't always work. I hate to say broken, but it was a D- for me because it was so predictable.

ORIGINAL: ralphtrick

ORIGINAL: jmlima

ORIGINAL: ralphtrick
I guess I should have mentioned that other than the cryptic mention in the what's new...

The basic issue is that the AI in COW was broken, so a lot of designers didn't bother spending a lot of time on the AI. It's also a moving target and changes some every release.
...

Uhhh???

So, what you're saying is that the game was broken if you intended to play vs the AI?...

Try playing COW again. Elmer is a limp dishrag. If you are next to an objective he ignores you. If take an objective behind him, he goes running back. To get around these issues, the designers tried plastering the map with objectives and doing other things that didn't always work. I hate to say broken, but it was a D- for me because it was so predictable.

Not arguing that.

But if it was broken it's one thing, if it was bad it's another thing.

I must confess I still have to find an even decent AI, at best what I found was a decent PO, and even those ones are predictable as hell, and make silly mistakes. Most of the times they are half decent tacticaly, but useless operationally, and even worst strategically.

If you want to post or email a save file where he is not doing what he should, I'll look into it. I remember flip flopping between waiting for help and tring to break out, but I don't remember where I ended up.

I think that if you use XML, i think you can force the ai to be a little more aggressive than you can in the editor, I'll check on that and get back to you.

The thresholds are to try to prevent turn burn. He starts wanting odds of at least 3 to 1 at the beginning of a turn and near the end will do anything better than 1 to 1. If you are expected to only have 2 sets of combat per turn, he probably isn't going to attack as much as he could. This is meant to overcome that problem, and make him attack more aggressively, trading ground for casualties, but risking early turn ending more often.

Ralph,

I have been doing a fair bit with the AI files but then I seem to be in the minority here in working with PO Assist. That facility is the best help scenario designers have ever been given for programming Elmer. I am hoping that designers using PO Assist to fine tune Elmer will lead to much better PO programming in the future.

The whole question of the AI is fraught with complexities. Given that designers have complete freedom to design any map they like, create any oob they like, create any TO&E and equipment they like, it always amazes me that Elmer does a fairly reasonable job. Some maps are PO unfriendly as are some oobs. Add in some dodgy PO programming (just plonk a load of objetcives around and set the strategic bias to berserk) and it's anarchy city!

Oh, and totally off thread, or possibly in thread but with a wobbly, I most definitely have seen supply units launch limited attacks. There is also something wrong with the reorganization code. Sometimes, instead of a computer controlled formation going into reorg, Elmer will set a formation to er.. actually I forget - I think its static but with a turn 1-4 delay and usually to activate on turn 1. Can't see this on PO turns unless you interrupt the PO and examine orange-banded formations but it's easier to see with PO Assist.

Not sure that I explained that last bit very well. Try Barbarossa 41, fast forward seven or eight turns on PO v PO and then interrupt the computer. Use PO Assist to play the Soviet side against an Axis PO and inspect all the orange-banded formations. Some will be formations in reorg while others will be (whatever) and activate on turn 1.

A developer's work is never finished!

Best wishes,
Steve

Thanks Steve, I'll give that a try.

ORIGINAL: jmlima
Not arguing that.

But if it was broken it's one thing, if it was bad it's another thing.

I must confess I still have to find an even decent AI, at best what I found was a decent PO, and even those ones are predictable as hell, and make silly mistakes. Most of the times they are half decent tacticaly, but useless operationally, and even worst strategically.

I'm not sure that the distinction means a lot, but there were a few logic bugs that meant it didn't perform even as well as it should have in CoW.

No arguments there, it's simple to program at the lowest levels, but the higher level the decisions go, the more challenging it becomes to even figure out what a 'good' decision is. Figuring out whether to go left, right, center, hold or advance becomes very, very hard. Once you throw in air, navy, embarking, airmobile and airborne, it becomes even worse.

Games like TOAW have it even worse because 1) the AI needs to be able to play close to historically in many cases, which means it's got to be constrained. 2) people can replay the scenario, so they are going to learn how the AI plays and learn how to defeat it.

That's part of the reason for AI files, if people can alter the AI for their opponent, there's a chance that they'll have a better experience than they do playing the default opponent.

I'm hoping that some people will post new AI files for popular scenarios for others to play against.

I think this is the sort of thing that Ralph is suggesting.

Two ai files (plus instructions) for Norm's Korea 50-51 scenario. The primarily active ai file has a radically altered Communist PO.

Best wishes,
Steve



PS. Well, forget that! I don't seem to be able to upload them. Hang on while I try to work this out. Can I not upload zips here?

Steve, rename the file extension to .txt, that should do the trick.

kLiNk, Oberst

ORIGINAL: Oberst_Klink

Steve, rename the file extension to .txt, that should do the trick.

kLiNk, Oberst

Ok. Here goes....

Download and rename to .zip

That seems to work ok. Many thanks.


No, that doesn't work either...

You need to right click on the attachment and then select save target as...

Bloody computers - I hate them. [;)]

Best wishes,
Steve