72 replies to this topic

[LarkinVB]

I think we're now ready for release.

I agree. Good job, Arkhan.

[thudo]

We haven't been known for "hotfixes" per se although Larkin is most gracious enough to publish them quickly.

We're always influx and improving the AI.

v1.6 is now on its run for a pre-Christmas release.

[Excedrin]

I tried 1.6b7 and noticed that it still captures criticals before regular strat points, so I made some minor modifications in captureplan.ai. My changed version is here:
http://dow.lerp.com/captureplan.ai

Please try it and integrate the changes if it works well.

Note that I've only tested it on Fallen City. Also, if there was some way to figure out if a point is a relic, that would be great (so it would cap the relic after strategic points).

Regarding playing against it, it seems improved. It did a few things that aren't optimal, like running to attack, then running away to defend (it should have sent "enough" to defend while maintaining its pressure). Also, it doesn't seem to understand harassment or splitting the army at all. Maybe some future version... The only other thing I noticed was raptors jumping into the middle of my army without support. It seems like the defendplan has no concept of "hold back until the defense will be successful." But I'm not really sure if that's a good idea, it's just how I play.

In other news, I've been thinking that it's possible to make an "unscripted" AI by coding an eval function and min-max search (like a chess program). It might be too slow in LUA to actually use it, but if I have some time I might work on coding something like this up.

[ArkhanTheBlack]

I tried 1.6b7 and noticed that it still captures criticals before regular strat points, so I made some minor modifications in captureplan.ai. My changed version is here:

In principle, the final 1.6 is out, but if Thud didn't finish the installer package yet, he might add it to the final.

Also, if there was some way to figure out if a point is a relic, that would be great

I don't think it's possible, at least I didn't find a way to do it.

Also, it doesn't seem to understand harassment or splitting the army at all.

yes, I think so too.

The only other thing I noticed was raptors jumping into the middle of my army without support. It seems like the defendplan has no concept of "hold back until the defense will be successful."

The infamous defend behaviour... That's another story for 1.7.

In other news, I've been thinking that it's possible to make an "unscripted" AI by coding an eval function and min-max search (like a chess program). It might be too slow in LUA to actually use it, but if I have some time I might work on coding something like this up.

The attack strategy works a bit like that now. It's not 'scripted' anymore. To decide all AI actions by some kind of chess field abstraction, you'll probably run into problems because of the lack of terrain information, performance issues and the complexity of the situation analysis. Not to mention the effort of rewriting all the strategy, plan and tactic code. Don't underestimate this. You'll need a lot of time and patience for this. Good luck if you want to try it!

[thudo]

I was about to send if off to packaging but I'll intergrate this captureplan and do a quick test then its good to go.

Hey.. at least v1.6 will be well in advance of Christmas!

Also, if there was some way to figure out if a point is a relic, that would be great

Thats not possible as I recall.. there is no code for the AI to acknowledge Relics. Thats something Relic will need to address..

Excedrin.. once more THANKS A TON for helping out! Any coding ideas you have are greatly appreciated!

Edited by thudo, 13 December 2005 - 03:23 PM.

[LarkinVB]

Good idea with strat before critical. I will further enhance the function so AI will prefer capturing strats which are nearer to enemy than those behind base unless the distance difference is too big.

Note that I've only tested it on Fallen City. Also, if there was some way to figure out if a point is a relic, that would be great (so it would cap the relic after strategic points).

Not possible accoring to BigFoot in march.

[Quietdeath]

If you release 1.6 on the relic forums I would include a "Known Issues" section so there are fewer people who say "I found bad behaviour X"
So, it is easier to concentrate on other aspects of the ai if you know what has already been discovered.

Another thing I want to add is that I often play against the AI and see things which can be improved but I don't know wether it is already fixed in a newer beta. So how about intern Snapshots which are not released and only submitted for users to test them without being in the beta test team.
If this is not possible - maybe you could post some replays more often, which interested users may analyze. (thanks to Excedrin who uploaded several)

*thumbs up* for your work and the release

[thudo]

Perhaps QuietDeath you want to be on the beta team for v1.7? v1.6 is now complete and we'll be concentrating on v1.7 before and after Christmas. The changelog for all our AI builds is posted on our site so its not for the public.

[ArkhanTheBlack]

I strongly agree with the known issues part. It should at least contain that the AI defense and ally support is not always optimal and that the take and hold code is far from being perfect, too.

we'll be concentrating on v1.7 before and after Christmas.

For me it's definitely AFTER christmas. I switched to XMAS mode sunday afternoon.

[thudo]

Yeaaa.. New Years it! >BurP!<

[LarkinVB]

There will be a 1.61 quickfix and I made some suggestions. It is NOT available yet.