Civ AI Survivor: Season Five

[thestick]

One of these days I'm going to go mad and quantitatively model this thing... hopefully before the next season.

Stupid idea is to run a multiple regression of AI traits on their power ratings to find the best AI traits. Stupider idea is to model the Civ 4 game as a rational choicemaker from microeconomics, who chooses its 'prefered' AI (winning the game). Then model which AI wins as a rational choice plus random error, and try to determine the preference relation that the game has over all AIs (e.g. which AI will win, given an arbitrary set of AIs).

Stupid idea is doable, stupider idea is interesting and will probably fall apart immediately.

[Fluffball]

It would be neat if there were a way to quickly run say, 10,000 games on random maps so we could get some real numbers for the AIs.

Anyone have a supercomputer handy?

[Sullla]

Game One is in the books! Thanks to so many people for turning out today; we had our most contest entries ever at 238 and peaked at 180 viewers on the stream concurrently at one point. Every time that we do this event it seems to get bigger and draw in more people watching. Here's a batch of links for the next game:

Main Season Five Webpage

Sullla Twitch Page

Sullla Discord Channel

AI Reference Guide from CivFanatics

Game One Livestream

Next Game: Opening Round Game Two

Schedule: Scheduled for Friday, 05 June 2020 at Noon EST

Game Two Video Preview

Game Two Written Preview

Game Two Picking Contest Entry Form

Congratulations to CymbalXL, Amicalola, eyser24, and ThreeLeggedChicken as the winners of Game One's picking contest with a very impressive 26 shared points. Scoring was strong across the board in this game with an average score of 10.63 points. We are continuing with our normal Friday date for next week's game. This time we're looking at Catherine, Suryavarman, Sitting Bull, and Wang Kon along with several other less notable AI leaders. Who do you have this week?

[Fluffball]

Minor request for the previews: Since this season has enormous swaths of tundra and ice, could you include a screenshot showing the resources of only that region? It's hard to tell if the southron lords have room to stick lots of seafood, silver, fur and deer cities in, or if the land is entirely useless. Thats a LOT of land in the south and if there's resources and hills to mine, farmable tundra, etc, we could be talking like an extra 10, 15 reasonable cities. For example DeGualle made great use of his ice (for whatever that got him), while Mansa founded no cities in the ice at all.

In past seasons it hasn't made sense to show resources because any green land is good for the AI, but ice is a bit different.

[GeneralKilCavalry]

One of these days I'm going to go mad and quantitatively model this thing... hopefully before the next season.

Stupid idea is to run a multiple regression of AI traits on their power ratings to find the best AI traits. Stupider idea is to model the Civ 4 game as a rational choicemaker from microeconomics, who chooses its 'prefered' AI (winning the game). Then model which AI wins as a rational choice plus random error, and try to determine the preference relation that the game has over all AIs (e.g. which AI will win, given an arbitrary set of AIs).

Stupid idea is doable, stupider idea is interesting and will probably fall apart immediately.

It would be neat if there were a way to quickly run say, 10,000 games on random maps so we could get some real numbers for the AIs.

Anyone have a supercomputer handy?

I actually do have a supercomputer handy, although I'm sure the department of energy would not be very happy with me installing games and forcing the ai to play against each other. Running the regressions or any MCMC would probably be fine.

With the debug console in-game, this is probably very possible:
Game.AIPlay can have the ai play on its own till some number of turns, and then gc can be used to print out the game context, units, research, etc. This would provide more than enough data for analysis.

The tricky part would be setting this up with some sort of macro to run the 10000 sims using Game.AIPlay and be able to extract the data from gc, but this is all solvable.

[thestick]

Thanks, I'll look into that, eventually, maybe.

[Sullla]

Written summary for Game One has been finished: http://www.sullla.com/Civ4/civ4survivor5-1.html

This game ended up being pretty straightforward after the early events played out. There's a link to the Livestream recording earlier in this thread and the YouTube version of the same recording should be up tomorrow. Good luck with the predictions for Game Two - I'll add the community summary for the next game on Wednesday.

[haphazard1]

Thanks for running this highly entertaining event again, Sullla!

I will second Fluffball's request for a bit more info about the icy areas. With these climate settings there is quite a bit of land in the frozen regions, and whether there are enough resources for useful cities can make a substantial difference. A zoomed out globe-level shot with resource markers turned on would help a lot.

[Charriu]

I would love to have this "globe-level shot with resource markers turned on" for all future AI Survivor games.

[Kuro]

Two questions

1. Can we get a link to the spreadsheet (or a copy of it) containing the current scores? I recall we had this before and would enjoy being able to do so again.

2. Given the AP was removed, was the suggestion of setting Espionage needed very low also done? It is particularly relevant for this map since Sitting Bull has high espionage ratings, so he might be more affected by this than most.