March Madness: Civ4 AI Edition

[Azza]

Hmmm. There's some interesting ideas here. Let me provide my own thoughts on some of the suggestions.

I'm surprised that several of the comments were opposed to having a balanced map for the final game. I heard repeatedly throughout the playoff games that each map was unfair for one leader or another. The first game was unfair for Huayna Capac, the second game was too easy for Mansa Musa, the third game was unfair for Boudica, etc. I thought that we'd use a single balanced map for the championship so that no one could complain that one or another leader had an advantage. I'm genuinely surprised to be hearing suggestions that this was a bad idea, never expected that. But if we'd used another random map, wouldn't there just have been more complaints that Leader X had too big of an advantage over Leader Y?

To me, it felt like it changed the rules for the last game. Seeing how different map shapes affected the way the AI interacted was very interesting to me, particularly the big U shaped map where the AI kept sending their stacks to the other side of the world instead of attacking their neighbours. Minor advantages to some civs over others due to the map was just part of the luck of the draw, and as we saw notably with Ragnar a raging pre-game favourite doesn't necessarily win.

In the grand scheme of things, a pretty minor nitpicky complaint, but it was done so well overall that it was the only issue I had really.

[Old Harry]

I'd like to see the AIs on a lower difficulty, with the Deity bonii it doesn't feel like they are playing the same game that we do. Also it'd be really interesting to use one well balanced map (perhaps using novice's tool) for all the games and to see how the different AIs use it over the tournament.

[Lord Parkin]

I thought the game settings were pretty much fine. You could turn off Aggressive AI next time - I really don't think it makes much of a difference (I did a lot of test games with it off and didn't notice any change in number of wars or survivors). Other than that I'd keep it the same.

Mapwise I didn't have much of a problem - some leaders definitely got a bit screwed over on some of the maps, but I know it's hard to find fair ones. Besides, it was fun seeing different landforms and guessing how they would play out. I think the final match being on a donut map was a bit of a break from the previous theme, but it seemed fitting for a more serious game so I didn't mind. I definitely don't think donut maps for the whole competition would be a good idea - while the Pangaea maps are inherently imbalanced, they make the games more interesting and varied. Thirteen games all on donut maps would be a bit too monotonous. And it's not as if this is a serious competition where we're trying to control for every variable and keep everything balanced - it's mostly about having fun.

I was actually thinking that there would be more suggestions on how to manage the scoring categories when I asked. I basically made them up on the spot when organizing this thing, and I'm not convinced that they were the best ones. Is there something else we could ask people to predict that would be more interesting? (Something that doesn't require tons of work for the one spectating the games, nothing like "total number of wars" please!)

The scores for the existing categories seemed about right - could maybe adjust the range for the "victory turn" points down a bit, but that's about all.

Possible scoring categories to add:

"First Threesome" - Predict which leader will be first to found their third city (or second city if it's not Deity - though I like it being Deity). 1-2 points?
"Last Threesome" - Predict which leader will be last to found their third city. 1-2 points?
"First Aggressor" - Predict which leader declares the first war of the game. 2-3 points?
"First Blood" - Predict which leader takes the first city of the game (not necessarily the same as "First Aggressor"). 2-3 points?

Could also bet on which leader gets a critical early wonder (Pyramids/Zeus).

[RefSteel]

I'm surprised that several of the comments were opposed to having a balanced map for the final game. I heard repeatedly throughout the playoff games that each map was unfair for one leader or another.

I found this a bit surprising too, but I suspect that for some, it was more fun to predict the outcome of a game when there were more variables to work with: Could Catherine's strong traits and AI "strategy" overcome her neighbor problems? Could Temujin fight his way out of his iceball start? Would Izzy's central position enable her to spread her religion everywhere, or would it just mean more neighbors for her to hate, attack, and be killed by? There's definitely an argument for embracing the randomness here and just seeing what comes out.

Of course, that's just speculation; I didn't mind the balanced map myself. I liked the (more-)random ones too, even though I commented about their imbalances in my predictions: I noticed the differences in start qualities and mentioned them in the course of my thoughts about what might happen in each game, but my comments were only negative if you assume (as I do not) that all the starts should be equal in a tournament of this type! The maps for Playoff games 2 and 3 would be pretty awful for MP games, but these aren't MP games with human players! It's impossible to please everyone, but I certainly hope no one expected you to carefully balance every map on behalf of these AIs!

Also: Thanks again for running the tournament! It was a blast!

[Kuro]

I'm surprised that several of the comments were opposed to having a balanced map for the final game. I heard repeatedly throughout the playoff games that each map was unfair for one leader or another.

I found this a bit surprising too, but I suspect that for some, it was more fun to predict the outcome of a game when there were more variables to work with: Could Catherine's strong traits and AI "strategy" overcome her neighbor problems? Could Temujin fight his way out of his iceball start? Would Izzy's central position enable her to spread her religion everywhere, or would it just mean more neighbors for her to hate, attack, and be killed by? There's definitely an argument for embracing the randomness here and just seeing what comes out.

Of course, that's just speculation; I didn't mind the balanced map myself. I liked the (more-)random ones too, even though I commented about their imbalances in my predictions: I noticed the differences in start qualities and mentioned them in the course of my thoughts about what might happen in each game, but my comments were only negative if you assume (as I do not) that all the starts should be equal in a tournament of this type! The maps for Playoff games 2 and 3 would be pretty awful for MP games, but these aren't MP games with human players! It's impossible to please everyone, but I certainly hope no one expected you to carefully balance every map on behalf of these AIs!

Also: Thanks again for running the tournament! It was a blast!

The only problem I had with the donut map was that the rest of the playoffs all had stuff that obviously wasn't balanced and such, then the big final winner takes all match did. It felt kind of like changing the game up at the last minute, y'know? It wasn't even something I'd really say was a big complaint, just felt slightly odd.

[Molach]

scoring

I'd vote more points for getting leader right. (Most important prediction)
Also less points for victory date and no bonus for getting it exactly right. Say 5 points if you get it within 10 years and that's it.

[antisocialmunky]

I think it just goes to show you can't please everyone. I actually was expecting something special for the final game. Both approaches have pros and cons. With random maps, you get really imbalanced but some interesting thought problems. In the Donut Map, its really down to who is neighbors with who and who gets to the middle first. Which is also interesting but a little more routine.

I actually wouldn't mine if we rolled/had plako make a single balanced map that we run every game on (Maybe that 3v3v3v3 PB map). That way we can compare opening between AIs and settling patterns in a more controlled setting.

Actually something I would like to see are teamers. See if the AIs are decent at cooperating. Maybe do a 2v2v2v2 with teams consisting of an econ AI and an warmonger AI.

[TheSunIsDark]

I might try an Archipelago Tiny Islands one and see how the AIs cope . A massive modded 52-civ game might be fun! Ofc, the observer would be difficult to set up.

[Tyrmith]

I was pretty happy with the setup. I like the au natural maps and the final map being balenced. I think it worked well.

What might be interesting to do would be to group up the AI types (IE religious nuts, warmongers, the tech focused...) and do a tournament that way. Some of those groups (ie research/economy focused) might be rather boring, but watching the religious nuts and the warmongers go at it could be pretty interesting.

[Hesmyrr]

Regarding the scoring:

* While 15 pts for hitting the victory date exactly was fun, it's massive influx of scores for what is basically a lucky guess. It's basically equivalent to perfectly guessing winner, runner-up, and first to die! I would prefer it decreased from 15 pts to 8 pt.

* There should be bonus points for "jackpots". If I'm to throw out random numbers, 20 pts for guessing winner, runner-up, and first to die correct (+5 pt) and 35 pts for perfect prediction of everything excluding the victory date (+14 pt).

* I personally didn't like having to guess the results of Ro8 group without having seen the map first, which plays enormous role in one's prediction. It influenced people's subsequent predictions as well because it had such massive point value, meaning you had to root for the leaders that you picked over the ones you knew were likely to win after taking look at the map.

I know some liked it, but removing such predictions with incomplete information would be preferable for me - or putting it into separate category of its own maybe.


Regarding the game:

* While I agree island-centered scripts are not a good idea, maybe there's other custom scripts that could work well with the format, like Toruslands (ocean setting) etc.?

* Is there no other method to track the AI players playing besides having the lurker civ be present? You wouldn't have to worry about the effect of open space created by removal of lurker civ's original location anymore.

* Slightly lower difficulty setting might be interesting; some civs with good starting technology were screwed over with the Deity bonus etc.


This is all I can think up for now