Ack! I'm afraid RC2 won't be "the one"

[Dilvish]

I just realized I have been letting an AI bug slip past me for a while now -- a significant chunk of the mid-tier ship designs the AIs should be using were getting bypassed due to a mis-designation in how I organize them & they would stick with an underpowered design for too long. They would (correctly) report they were able to make all the right designs, they just wouldn't always choose the one they should during a particular phase of tech development. It's a large enough set of designs that it very well might make a noticeable difference in AI performance (though it's hard to say; the designs being passed over would have been more expensive to produce so it might all roughly be a wash). I just committed a fix.

[Karoushi]

I just realized I have been letting an AI bug slip past me for a while now -- a significant chunk of the mid-tier ship designs the AIs should be using were getting bypassed due to a mis-designation in how I organize them & they would stick with an underpowered design for too long. They would (correctly) report they were able to make all the right designs, they just wouldn't always choose the one they should during a particular phase of tech development. It's a large enough set of designs that it very well might make a noticeable difference in AI performance (though it's hard to say; the designs being passed over would have been more expensive to produce so it might all roughly be a wash). I just committed a fix.

Sounds good, my experience with the AI is that they are somewhat weak but I haven't been playing with adequate settings apparently so I am re-testing them.

My general experience with them is that they always had weaker ships then me so this fix sounds good.

[ShockPuppet]

mid-tier ship designs the AIs should be using were getting bypassed [...] I just committed a fix.

So this is how I die

[Geoff the Medio]

I'm wondering about the timing for an RC3 build. There have been some mixed reviews about AI performance / difficulty in the last couple builds, and some discussion about changing how AI personalities are selected given the galaxy setup setting. Would waiting until next Monday to put out RC3, with no non-trivial game content balance changes permitted in that time, be useful for tuning the AI and implementing the personality tweaks and optimizations?

[Dilvish]

I'm wondering about the timing for an RC3 build. There have been some mixed reviews about AI performance / difficulty in the last couple builds, and some discussion about changing how AI personalities are selected given the galaxy setup setting. Would waiting until next Monday to put out RC3, with no non-trivial game content balance changes permitted in that time, be useful for tuning the AI and implementing the personality tweaks and optimizations?

Hmm, well, "would be useful" is a fairly broad phrase; I guess it depends on what you're hoping to get. I just want to be careful about not getting expectations too high, whether I had a few extra days or not.

I'm very close to solving the problem with the AI sometimes reverting to that old behavior where it would stream ships into known death. Somewhere between getting its "flight path" and implementing it, or while implementing it, the penultimate system in the list is getting deleted / omitted. The AI fleet can still navigate to its destination, but my threat checking for that penultimate system is not getting a chance to operate. I'm pretty hopeful of getting that fixed soon today.

Shifting the AI personality selection from being a distribution to a single specified level for all AIs in the game is fairly trivial; please clarify if that's what you'd like me to do.

So that, with the fix that went in for AI ship designs, would be two to three AI fixes/mods of minor to midling significance since RC2; the personality selection may wind up being of midling to major import to player experience. It will almost surely still leave the AI competitiveness noticeably short of what it was a month or 2 ago; it seems the game dynamics have changed too much for me to catch up with simple tuning of tech priorities or Industry/Research relative priorities (though I do continue to experiment with such).

I kind of think I'd be inclined to get out an RC3 fairly quickly -- once I fix this pathing-into-death bug, and (assuming you want it) changing the AI aggression setting to a fixed level for all AIs in a given game; I expect the code could be finished within a matter of hours from now. Then we could get a few days of feedback on revised AI performance while I try to see if I can identify any other bugs dragging it down, or otherwise come up with any noticeable improvements to AI "tuning", that might make an RC 4 worthwhile instead of just releasing RC3.

I suspect it is going to take some more major work than that before I can again be truly pleased with AI performance, though.

[Karoushi]

Shifting the AI personality selection from being a distribution to a single specified level for all AIs in the game is fairly trivial; please clarify if that's what you'd like me to do.

What if the user was aloud to select the aggression level for all of the AI's individually? Kind of like (an example) Red Alert 2, where you can set some AI's to "Brutal" and others to "Turtle", others to "Rush" ~ that way the player can choose how many AI's will have which aggression level, instead of just facing a massive wave of all hard, or all soft targets.

The only issue I see with this is you'd have to add an area for the user to do this on the galaxy setup screen.
The way it works in Red alert 2 is kind of like the multiplayer lobby and all the users and ai's are treated as individual players into which you can set their colors, difficulty level, ally or "teams" (team 1, team 2, etc) ~ which would allow the user more control over their gaming experience with the AI.

@ post below ~ Sorry, my bad, just wanted to mention it while it was being discussed.

[eleazar]

What if the user was aloud to select the aggression level for all of the AI's individually?

That's beyond the scope of 0.4.2

[yandonman]

still leave the AI competitiveness noticeably short of what it was a month or 2 ago;

I don't think this should be a criteria for 0.4.2. AI improvements can (and most certainly will - hopefully ) be made as we march towards later versions.

The only thing I would take is the "flattening" (E's term) of the AI aggression galaxy setup option - if someone sets a AI aggression level, 75% or more of the AI's should be at that level (if not 100%).

My recommendation would be to get that fix in, spin a new RC3 ASAP (don't wait till Monday) and baring no major issues found (not expecting any) - done - declare 0.4.2.

[Dilvish]

OK, I fixed (or at least substantially reduced the frequency of) the AI bug with streaming ships into known death. I also narrowed the AI aggression distribution as per yandonman's suggestion -- so if you specify Maniacal as max, you'll get ~75% maniacal and ~25% Aggressive. Those changes are committed. I may still do a little work tonight on AI tech priorities.

[AndrewW]

OK, I fixed (or at least substantially reduced the frequency of) the AI bug with streaming ships into known death. I also narrowed the AI aggression distribution as per yandonman's suggestion -- so if you specify Maniacal as max, you'll get ~75% maniacal and ~25% Aggressive. Those changes are committed. I may still do a little work tonight on AI tech priorities.

Been playing with different AI settings on here, had one on Typical where it expanded less then the Turtle setting...

[Dilvish]

Been playing with different AI settings on here, had one on Typical where it expanded less then the Turtle setting...

I don't believe the changes I committed have yet been put out as a package; you'd have to compile from the source code for yourself.

[AndrewW]

Been playing with different AI settings on here, had one on Typical where it expanded less then the Turtle setting...

I don't believe the changes I committed have yet been put out as a package; you'd have to compile from the source code for yourself.

Oh I've been doing that actually... But haven't pulled down an update yet.

But was just meaning with the current rc2.

[Geoff the Medio]

In a server log just posted about unrelated problems, I note a bunch of these error messages:

Code: Select all

2013-02-13 03:02:19,991 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus FOCUS_INDUSTRY attempted to be set for object w/ dump string: ResourceCenter focus: 
2013-02-13 03:02:19,995 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus FOCUS_RESEARCH attempted to be set for object w/ dump string: ResourceCenter focus:
2013-02-13 03:02:20,004 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus  attempted to be set for object w/ dump string: ResourceCenter focus:

Perhaps this is a contributing factor to AI performance changes from previous versions?

[Dilvish]

In a server log just posted about unrelated problems, I note a bunch of these error messages:

Code: Select all

2013-02-13 03:02:19,991 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus FOCUS_INDUSTRY attempted to be set for object w/ dump string: ResourceCenter focus: 
2013-02-13 03:02:19,995 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus FOCUS_RESEARCH attempted to be set for object w/ dump string: ResourceCenter focus:
2013-02-13 03:02:20,004 ERROR Server : ResourceCenter::SetFocus Exploiter!-- unavailable focus  attempted to be set for object w/ dump string: ResourceCenter focus:

Perhaps this is a contributing factor to AI performance changes from previous versions?

No, that's been going on forever; it's generated when the AI is testing what target output it could get from setting either Industry or Research focus on the planet. One of these days I'll get it cleaned up to not generate those. It doesn't hurt the AI's planning, for now, though I suspect that when I start supporting Growth Focus or others I'll need to have it cleaned up.

The performance hit is largely because we've made a number of significant changes affecting how empires grow, which was where I had previously been able to give the AI relatively strong skills -- we've vastly increased the costs of colonies and outposts, and almost entirely done away with what had been a very significant, easily plannable, output from planets with the Lithic specials, along with a number of other changes. But those two are probably the biggest ones impacting the AIs.

I'm working on adjusting the AIs research priorities and colonization strategies, but this slower growth of empire production puts a much greater premium on war tactics, which the AI was never great at. I'm proud of the improvements I've been able to make in that respect from the point where I took it up, but there is still a long long way to go.

**edit** Now that we have the Order process checking to prevent improper Focus change orders from taking effect, we should probably do away with that Server message as it clutters the log and is generally unneeded (only serving as a reminder to me to reorganize my AI code, hehe)

[Dilvish]

OK, I submitted another round of AI updates to SVN (5763). It would help me if those of you with compile capability could give that a run-through and give some feedback.