Are you, in fact, a pregnant lady who lives in the apartment next door to Superdeath's parents? - Commodore

Create an account  

 
Civ4 AI Survivor: Season Four

(June 11th, 2019, 11:59)Zalson Wrote: I am excited to announce that through the first round of the competition, Fatkid, random.org, and I are all in contention for the coveted Germany award: a participation award for those of use who fail completely. The three of us are neck-in-neck at a total score of 45 points. That's an average of 5.63 points per game!

Why a Germany award?

The other Civs with a score of 0 in the power ranking have given us wonders like Sitting Out Bull, Hatsheput of North Korea, Victoria's Nuclear Exchange, and Washington's Apostolic Cheese Survival. Nothing can compare to the ineptitude of the German civilization, who also dazzled us this season by ... dying second in both their games. They're honestly not even worth picking for first to die any more!

I'd like to thank my wife and children for their encouragement as I attempt to fail my way into the annals of history.

As spokesperson for random.org in this competition, I have been directed to say that the dice are very thankful for the efforts of quantum in this near singular achievement. Now to choke on my fellow German's dust and win sole hold of the coveted prize.
Travelling on a mote of dust, suspended in a sunbeam.
Reply

Re Industrious trait: There are SO many factors it's hard to say what trait is good versus what uniques are good and what flavors are good.

You could do a Masters thesis on this. It would take a huge amount of work, but you could "Moneyball" every single aspect. I mean off the top of my head you can sort of quantify the aggressive trait into something like a 4% (give or take 100%) bonus to the hammers put into all things military. After you quantify the easy stuff I'd move on to quantifying how each flavor influences things, like industrious and a production flavor would get forges up faster, thus getting everything else up faster.

You'd have to quantify peace weight and religion as well, as I think pretty much everyone agrees Mansa is far and away the best leader including everything (his two uniques are excellent, flavors, traits, etc.) but his peace weight screws him.

I honestly think it would be a really interesting grad school thesis for someone interested in predicting things that seemingly can only be "felt", exactly like Moneyball.
Reply

It is actually kind of impressive for random.org to do this badly. lol
Reply

Fluffball, I agree there is some interesting work that could be done with the results of the four seasons of AI survivor. It would be fun to make a 'super AI' by picking a leader and civ and then modifying their XML to have an optimized combination of factors. Obviously this would be optimizing for the conditions of AI Survivor, but it would be interesting to see how such an AI might compete. Give one leader the 'best' traits, research flavors, unit build tendency, aggression, wonder preference, religious preference, peace weight, ability to declare war at pleased, etc. How well would it perform?
Reply

(June 11th, 2019, 21:45)haphazard1 Wrote: Fluffball, I agree there is some interesting work that could be done with the results of the four seasons of AI survivor. It would be fun to make a 'super AI' by picking a leader and civ and then modifying their XML to have an optimized combination of factors. Obviously this would be optimizing for the conditions of AI Survivor, but it would be interesting to see how such an AI might compete. Give one leader the 'best' traits, research flavors, unit build tendency, aggression, wonder preference, religious preference, peace weight, ability to declare war at pleased, etc. How well would it perform?

For the past couple years I've stopped playing the typical civ4 and in an effort to reinvigorate (VERY successfully!) the game I've been teaming up with AI. 2v4v4 or whatever the heck seems interesting. 

When you team with an AI you get to control their research (via asking them) and their diplomacy (just the default). Let me tell you that as a teammate King Musa blows the competition so far out of the water it's not even funny. No one even comes CLOSE, not even Justinian or Capac. When I beeline Musa to rifling, when I tell him to declare war on a weak friendly neighbor, it's just so far ahead of anyone else it's ridiculous. He often has the highest power rating by *magnitudes* over people like Caesar or Napoleon or whoever, even on relatively high difficulties (no way I'm playing diety for fun.)

Mansa is basically the super AI if you account for his minor weaknesses. If he had a low peace weight AI survivor wouldn't even be a competition.


Edit: These team games also put AIs in the same situations so you can compare them. For example if you had a team of Julius and Musa, Julius is likely going to be much weaker than Musa because part of Julius' strength lies in opportunistic bullying. He under performs in team games where the human controls the diplomacy. Pacal, as a teammate.... you might as well have had the barbs. Pacal is a colossal failure in every team game I've ever used him, and since I realized that I've never picked him in Survivor. Pacal's AI is just plain wrong.
Reply

It is a difficult balancing act, trying to tune an AI for optimal performance. You could make Mansa more likely to survive by giving him a higher unit build preference and more military research preference, but then he would not be as effective at being the economic powerhouse Mansa we know.

Something like Civ AI Survivor would ideally be part of the initial game/AI development process, with the developers using the results to improve all of their AIs.

On Pacal, I agree he is often disappointing. Such excellent traits, but he just does not seem able to put together a strong game most of the time.
Reply

Industrious trait in general is one of the hardest to analyze as other have mentioned. The wonders game is zero sum so the effectiveness of industrious is inversely proportional to the amount of AIs with industrious in a game.
In Soviet Russia, Civilization Micros You!

"Right, as the world goes, is only in question between equals in power, while the strong do what they can and the weak suffer what they must."
“I have never understood why it is "greed" to want to keep the money you have earned but not greed to want to take somebody else's money.”
Reply

We could just do a silly curve-fitting first cut and multiple regression a bunch of AI characteristics on power rankings dancing. This is horribly unrigorous since the AI power scores are not independent (there's essentially only a finite number of power points divided among each of the AIs), but I don't know enough statistics to figure out how to fix that.

Then we use some variation of k nearest neighbors or whatever to predict games. Define AI characteristics (mean and standard deviation of aggression, etc), choose the nearest game to that, and use that to predict the results. Pick the AIs who have the closest profile to the AIs from the other game as the predicted winners/first to die, and just copy over the victory conditions/dates pimp. Or we run a bunch of multinomial regressions  smoke.
More people have been to Berlin than I have.
Reply

Great discussion in this thread, I really enjoyed the look at the production focuses of the AIs and how that impacts their performance. Thank you very much for putting that together Borsche! thumbsup I'll have to think about doing a deeper dive into some of those aspects of their programming after this current season wraps up.

Community predictions have been added to the preview page for the Wildcard Game! Here's a quick summary:

[Image: survivor4-9-8.jpg]
[Image: survivor4-9-9.jpg]
[Image: survivor4-9-10.jpg]
[Image: survivor4-9-11.jpg]

The community vote picked Mao as a strong favorite for this game, with more than half of the contest entries selecting him to be the winner. Suryavarman was a clear, if distant, second in this category and Wang Kon also drew a few votes from his fans. The other three leaders had virtually no support from the community and had to be viewed as longshots. The Runner Up category indicated a preference for the same three leaders in some combination, either Suryavarman or Mao or Wang Kon. At the other end of the spectrum, Wang Kon appeared again as a slight favorite to be First to Die, with Ragnar hot on his heels in this shameful category. Amazingly, more people picked "None" in this category (1) than picked Mao (0)! If anything confirmed Mao as the favorite, it was that.

Finally, here are some of the best/craziest written predictions about what would take place during the game. There were many other excellent entries but I had to pick and choose my favorites to keep this from running on too long. Thanks again for the submissions!

Keler (artdeco): I know one thing for sure that Wang Kon will not win this game no matter what. And I am not a fan of Suryavarman the Clastrophobie. Creative so what? Just like Pericles in the last game I guess. I am a fan of Saladin. He is the best AI here, but his starting location kind of sucks. Beside the community here thinks he is bad AI based on 3 seasons only. Time to prove them wrong Saladin! Saladin's starting position seems the worst here but the map looks balanced enough except Mao Zedong's outstanding starting location. So I guess Saladin will just try to out tech, something he is good at and eventually capture few cities from Khmer and Mao Zedong will just snowball over Korea and dominate the map.

Lone Wolf: No leader here really stands out... my prediction is that financial Wang Kon will win - he certainly has the capability of winning - after Ragnar conquers the Khmer and divides the Chinese with Wang. Saladin will just turtle up in his corner of the map, fending off conquest attempts but not doing much else. Optionally, he may conquer the English.

Warclam: Mao's spot looks good. Great land, Ragnar and Suryavarman aren't going to be inclined to mess with him. Now, peace weight suggests Wang will die first. Even Saladin is 4 away from him. On the other hand… not only will the Troll Force protect Wang, but Suryavarman is in the middle of the map, so his lack of protective is going to get him in trouble with the raging barbs. I feel like Suryavarman will be kept weak by barbs, then taken down first, to Saladin's benefit. Meanwhile, Mao techs like crazy, joins an almighty dogpile on Wang, then devours Churchill. Ragnar flails like a ninny until he picks a fight with Saladin and gets spanked. Mao gets along well enough with the northerners that it probably won't be domination, so… space it is.

Hydrocyanide: These leaders are mostly terrible, so here's just hoping for Wang Kon Magic.

RefSteel: Were the south-central AI any other high-peace-weight leader, this game could be expected to play out by the (little red) book. The northern warmongers would band together against their most-hated enemy, allowing China, starting closest to the spoils in the first place, to join in decisively at the end, wind up with most or all of the victim's land, and snowball from there. But this isn't just any high-peace-weight leader: The question of the game can only be, "How is Wang Kon going to get out of THIS one?" My answer? *I have no idea!* Not because it seems impossible, but because I've long since given up any attempt to predict the acknowledged lord and king of all the trolls.

Dantski: Mao and Ragnar bully Wang like a couple of power mad moderators punishing a troll for misbehaving. Mao then hoovers up the map counter clockwise before teching to space, taking his good buddy Ragnar with him. Space Vikings become a thing.

Latif: Well, Mao looks like he has the best land - lots of floodplains, and a private peninsula to expand into. I had picked Ragnar to finish second last game... won't make that mistake again! He's too psycho for his own good, and it looks like his start will be slow. Rags will probably declare an early war, waste lots of units, and start falling behind in tech as well as power as he throws his units away against city walls. I have Ragnar first to die, and that benefits Mao. I see Mao jumping out to a quick start and an early score lead, if he can handle the barbs as well as Ragnar's possible aggression. Mao's other neighbor Wang Kon has kind of a crappy starting position, one off the coast, not many rivers, no seafood in sight on his (tundra) coastline. Second place is hard to call, but with the logic above of Ragnar self-destructing, Sury, his other neighbor, also stands to benefit. He could very well have peace as well with Saladin who will probably spread his religion to him. Mao by domination, thought about culture because of the Pavilion, but his disdain for religion probably rules that out, and with his weak neighbors he could easily get really big.

Commodore: I can't overemphasize how scared I am of my prediction: Wang Kon first to die, Mao and Sury profit the most to go to the playoffs. This is directly challenging Troll Kon to troll me personally...and Ragnar the Random is here to help him.

LinkMarioSamus: I'm going to fall back on my metric of past competition these leaders have faced that served me so well in picking Tokugawa to win that one game and hasn't done much for me since. I do think it helps explain some of the results of this season, but who even knows anymore? I guess it's as good a metric as anything else, and it predicts Wang Kon to win with either Churchill or Saladin taking 2nd - going with the former because he's more likely to be allied with Wang Kon - and Suryavarman dying first. The reason for Wang and Churchill is because they were both in the one Season Two game with Huayna Capac, Boudica, and Stalin, and the only real dud AI leader in that field was Montezuma. Wang Kon was also in the Season Three game with Charlemagne, Huayna Capac, and Tokugawa. Maybe this all seems arbitrary, but again, who the heck knows in this crazy season? To be fair I could easily see Wang Kon and Ragnar both attack Suryavarman early and possibly work together to kill him or at least wear him down in concert. Then maybe Mao is next on the chopping block, and Saladin doesn't succeed in spreading his religion very well? Again, who in the world knows this season. Predicting these games can be a bit of an exercise in futility, but we're largely doing this for fun so I don't have much of a problem with it.

TMIT: Mao and Sury are the favorites here, and I like Sury a bit more. I'm gambling that Ragnar doesn't target him first, but this is not a fast-expanding AI field other than Sury so he can jump out to a big city count fast. I expect Saladin will found a religion he shares with Sury and mostly get left alone. Churchill will be unpopular on this map and sucks in general so he's my vote for first to die. Mao is #2 by default but could easily be #1, his starting position is fine too and he has a good history in AI survivor. Ragnar takes the usual kingmaker role based on who he attacks. He almost never makes himself king, he's generous like that. Generous with units.

Shpoko: The Troll King will have to be at the top of his game to make it through this one, I just don't see it happening. Mao and Surya team up to take out Wang and Churchill, then race to space.

Faded_Outline: I can't parse this... raging barbs throw any predicted game plan into doubt, and with four protective civs in play who knows who, if anyone, is going to feel like expanding. In the midst of the chaos, Wang Kon is laughing at us. God help our scores crazyeye

Wildcard Game Picking Contest Entry Form
Follow Sullla: Website | YouTube | Livestream | Twitter | Discord
Reply

Sullla, your preview link is to the game 8 preview rather than the wildcard preview.

This one should be interesting, with this cast of characters. Will we see more troll magic? lol
Reply



Forum Jump: