SMAC 444 (AI experiment)

[kyrub]

Secret projects have yet another rush cost formula.

Yes, that's 2* 2*Mm and 4* 2*Mm for Voice of Planet and Ascent to Transcendence.

The bug stretches over here: the AIs pay 2* (2*Mm + Mm*Mm/20). There are other energy leaks in the same routine. AI never checks if it buys at least one mineral, so it can silently waste energy turn after turn with no benefit. This really happens, especially with SPs. I guess it would be easier to forbid AI to spend on SPs. This could mean a lot more energy credits for other things - like upgrading units and hurrying facilities.

[bdanv]

This really happens, especially with SPs. I guess it would be easier to forbid AI to spend on SPs. This could mean a lot more energy credits for other things - like upgrading units and hurrying facilities.

the AI does a poor job at building SP anyway. forbidding the AI to rush buy will handicap them further.
i get it you can't make them check all 3 formulae above (units,facilities,SPs)
maybe making them disband more crawlers and/or other units could help.

[kyrub]

I am not against AI being more aggressive when trying to get SP. Not at all!

The problem is: AI uses Energy_reserves/50 every turn for SPs, but it needs at least 4 EC investment to get at least 1 mineral / turn. Otherwise the investment is totally wasted. So if AI has < 200 EC (which is quite often as I've seen), it keeps wasting its resources without any effect. About 100-200 EC per Secret project - you get the idea. I guess it's a copy from Civ series, horrible.

[kyrub]

New version: SMAC_444_k

Download: http://www.mediafire.com/download.php?6redidk4z42h1mo


Features (mostly bug corrections)

k1 stockpile energy bug
k2 swap base bug
k3 children creche bug
k4 AI hurrying price bug
k5 AI can build seaformers earlier
k6 more colony pods for University and Morgan
k7 reversal of distant AI cities (AI can build cities at closest distance again)

[kittenOFchaos]

Thank you for all your hard (enjoyable?) work.

I'll see if I can give it a try this weekend and let you know what I think.

[kittenOFchaos]

A few comments.

I played as Yang for a decent bit.

The AI are terraforming much better, very impressive. If you remove the penalty mines cause farms the AI mine+farm approach will be highly effective.

Morgan and University are also expanding much, much better which makes them decent - a vast improvement. In the game I played, the Gaians however weren't expanding - have you touched that aspect?

So basically, so far so good

[Petek]

New version: SMAC_444_k

Download: http://www.mediafire.com/download.php?6redidk4z42h1mo


Features (mostly bug corrections)

k1 stockpile energy bug
k2 swap base bug
k3 children creche bug
k4 AI hurrying price bug
k5 AI can build seaformers earlier
k6 more colony pods for University and Morgan
k7 reversal of distant AI cities (AI can build cities at closest distance again)

kyrub,

I tested the stockpile energy bug fix. Can report that both parts of the bug (as stated in post #39) are fixed. What are k4 and k5? Thanks.

[kyrub]

Morgan and University are also expanding much, much better which makes them decent - a vast improvement. In the game I played, the Gaians however weren't expanding - have you touched that aspect?

No, Gaians should be fine - they were AI winners in one of my games. Maybe it was a bad start for them, an island or very near another player? Also, some of the factions are programmed to take very risky movement (farmer's gambit) very early. Gaians could belong in the cathegory. They can be swiped badly by worms, as I've noticed already. Last possibility: there's another quite serious bug in AI settlement, the colony pods seem to be redirected every turn and they may move in circles endlessly. This happens many times a game. I haven't found the source of the bug yet.

@Petek
Thanks for stockpile testing, I had no time to do it myself.
k4 - see post #44. The AI used the same prices for hurrying (speed building) units and buildings, while the player normally has much better prices for buildings than for units.

An example for k4:
Your tree farm needs 40 units to be finished. You pay Mm * 2 = 40 * 2 = 80.
AI pays Mm* 2 + (Mm*Mm) / 20 = 40 * 2 + 1600 / 20 = 160.

You see the difference. The same bug was applied to Secret Projects building (huge prices). The AI should now be more competitive in this regard.


Sea formers (k5):
Before, AI waited until first sea base is founded to produce a sea former. Now, it will try to build a few seaformers before to speed the land bases' development.

[kittenOFchaos]

Kyrub, I've played another game and this time the Gaians are fine. Previously it must have just been a bad start, though it didn't look that way as they had plenty of land.

In this test game the Believers got the Monsoon Jungle and looked to be beating on everyone, but the Gaians and University together overwhelmed the Believers, it was quite impressive. However, the Peacekeepers this time expanded poorly and were almost destroyed but for my intervention. All the other factions were expanding quickly and providing a challenge though the Morganites aren't handling the unhappiness from units away from base in Free Market well.

I also noticed the Hive build a sea former, then they build two sea bases.

The changes made have made the AI far more competitive, that is for certain.

[kittenOFchaos]

What next Kyrub?

[kyrub]

Well, I am bundled with work in RL. And I am a lazy person. And this patch does not exactly make masses waiting, so it can proceed rather slowly.

(I have specifically set myself only two immediate goals:
- destroy the Talent facility bug, since that one used to drive me crazy
- learn AI to make better social engineering choices)

What next - your other suggestions?

[kittenOFchaos]

My wish is to see the AI able to build boreholes and echelon condensors.

[kyrub]

I am almost sure I can push AI to drill more boreholes. I think they already make more of them now, but it can be even better. Condensors, we will see.

I need to readdress the probes quantity. They are everywhere now.

AI movement... is an important chapter. Not only air power, but probes staying away from military units, colony pods, formers avoiding worms, units moving in stacks (this one seems quite hard). Probes defending. There's a big room for improvement. Although current state is not horrible. And I understand only about 2% of that part of the code now.


I have also small plans for AI diplomacy: Like making the Ask_for_loan for much bigger money, increasing the amount of cash you need to bribe someone to vote for you... If ever I can find AI-AI diplomacy, I'd like to make non-aggressive AI much more careful to stir a war.

[kittenOFchaos]

I'm so pleased about what you're finding possible, Kyrub.

It may seem quiet, but there are alot of people that still love this game. They're probably just not aware of what you're doing for this game.

I appreciate you

[kittenOFchaos]

Kyrub?