AI Survivor Switchheroo

[Thrasybulos]

Yes, I checked a few, and saw the same thing.
Would have to be cross-referenced with game composition: if there's one thing which can scuttle an overpowered starting position, it's a hostile diplomatic environment.

 

[Fippy]

So Hatty easily wins the "most robbed AI of all time" title?

 

[Thrasybulos]

She certainly hasn't been lucky: between "lamb to the wolves" setups, getting attributed dud starts, and getting "outlier-effected", she certainly seems to have offended the AI Survivor gods!

 

[Keler]

My Maps

Spoiler Map Name: Wildcard 6 :

Spoiler Map Name: Wildcard2 6 :

Spoiler Map Name: Wildcard 7 :

Spoiler Map Name: Wildcard 8 :

Spoiler Map Name: Wildcard 10 :

Spoiler Map Name: Wildcard2 10 :

Spoiler Map Name: Wildcard3 10 :

Spoiler Map Name: Wildcard 11 :

Spoiler Map Name: Wildcard 12 :

Spoiler Map Name: Wildcard 13 :

Spoiler Map Name: Wildcard 14 :

Spoiler Map Name: Wildcard 15 :

Spoiler Map Name: Wildcard 16 :

Spoiler Map Name: Wildcard 17 :

Spoiler Map Name: Final Circle :

 

[Keler]

add Napoleon and Suleiman to the Hatshepsut's list of unlucky leaders


I believe I presented everything I did. Feel free to request anything.I can post worldbuilder files etc. of my maps for anyone interested to look at them.
From now on I will update here every time Sulla makes a season.
To be continued with season 8 game files.

Also Thrasybulos that's a lot of work for just one map! I will wait what your conclusion will be like so we can find some consistency perhaps

 

[Saxo Grammaticus]

Interesting that even the really powerful starts don't seem to be all that dominant in this. Very few over 60% winrate.

And then you see Louis XIV had two favorable starts that set him up for a fight for the title. I am most familiar with the maps from Season 7, and I can also appreciate the starts where leaders did much better in the Alternate Histories than the map would suggest. In Game 6 Willem came out 45%/10% against the start's 14.3%/14.3%. Compare Darius' 45%/15% to the map's 42.9%/19%. Of course, the benchmark for this is probably Huayna Capac's ability to make the most of a bad start.

 

[ManiaMuse89]

add Napoleon and Suleiman to the Hatshepsut's list of unlucky leaders


I believe I presented everything I did. Feel free to request anything.I can post worldbuilder files etc. of my maps for anyone interested to look at them.
From now on I will update here every time Sulla makes a season.
To be continued with season 8 game files.

Also Thrasybulos that's a lot of work for just one map! I will wait what your conclusion will be like so we can find some consistency perhaps

Napoleon is one of the strongest leaders in my ranking too. Currently sitting at #3.

Hatty is very mid-tier for me, almost in the middle of my rankings. Certainly not a bad leader when she goes for quick cultural wins but Ramesses has been more consistent.

In my series I am doing completely random maps and random leader combinations though, my theory being that the stronger leaders should be able to deal with dud starts better than weak leaders. Although having looked at your results it looks like that might not be a true assumption. I would still always fancy HC over Frederick in the same starting position though.

 

[Fippy]

HC doesn't really play better than other AIs, but he has great traits & unique stuff.
And a peaceweight advantage against the most dangerous AIs.

Deity players would always vote for him as easiest leader to win with..
considering that, his performances are expected.
But not like he defends or attacks better than others

I never liked picking him (in Sulla's contest)..feels like picking the best package blindly (unless he gets great starting posis..which prolly results in boring games).

 

[Keler]

True, HC really get away with so many diplo bonuses on top of everything else, Cyrus or Mehmed with double power just won't declare at him... and then he starts to outtech.
"Huayna Capac is the best AI" is probably agreed on everywhere.

I really don't understand is what happened to Frederick to end up this bad. I thought he would at least have some more victories to be around Sitting Bull in the ranking, or just close to others.

Maybe compared to his allies Wang Kon,Roosevelt,Lincoln he must have it worse. Be it tech preferences flavour or trait combinations, this exact settings etc. I also wonder why Darius doing so great meanwhile Wang Kon is terrible. Because this is not the first time I made a ranking and there is some consistency between them Actually this is the biggest ranking of me so far. Older threads were like 500-600 games max. Who knows
It took almost 4 months for this one to happen.

In my series I am doing completely random maps and random leader combinations though, my theory being that the stronger leaders should be able to deal with dud starts better than weak leaders. Although having looked at your results it looks like that might not be a true assumption. I would still always fancy HC over Frederick in the same starting position though.

How many games you got so far that Napoleon is ranked 3? And is the map type callled RandomScriptMap?

 

[ManiaMuse89]

True, HC really get away with so many diplo bonuses on top of everything else, Cyrus or Mehmed with double power just won't declare at him... and then he starts to outtech.
"Huayna Capac is the best AI" is probably agreed on everywhere.

I really don't understand is what happened to Frederick to end up this bad. I thought he would at least have some more victories to be around Sitting Bull in the ranking, or just close to others.

Maybe compared to his allies Wang Kon,Roosevelt,Lincoln he must have it worse. Be it tech preferences flavour or trait combinations, this exact settings etc. I also wonder why Darius doing so great meanwhile Wang Kon is terrible. Because this is not the first time I made a ranking and there is some consistency between them Actually this is the biggest ranking of me so far. Older threads were like 500-600 games max. Who knows
It took almost 4 months for this one to happen.


How many games you got so far that Napoleon is ranked 3? And is the map type callled RandomScriptMap?

I have completed round 37 now (8 games per round involving all 52 leaders so 296 games played) which seems a bit obsessive but it is quite easy to run them in the background whilst doing something else. First goal is to have every leader play every other leader at least once although the random number generator hasn't quite made that happen yet.

Not enough games for anything to be hugely statistically significant but I am definitely seeing patterns with more successful and less successful leaders.

HC is up to 15 wins out of 37 games which is a pretty impressive winrate. Next up is Napoleon on 12 wins although he is 3rd in my scoring system which also takes into account games survived.

The only leader yet to score a single win is Frederick. I haven't really paid much attention to what he does or doesn't do to be so ineffective. I have a feeling that he just goes for really dumb early game tech decisions which slow him down a lot during the most important phase of the game. His survival rate is actually alright, there are a bunch of other leaders at the bottom of the pile who have much worse survival rates, but those leaders do seem to be able to win the occasional game when they manage to snowball.

The maps I use are normal size pangea, temperate climate, solid shoreline, normal sea level for 6 AI games, low sea level for 7 AI games. Other settings/starting techs/units the same as Sulla, I got rid of the AP in the XML I only re-roll maps if the 'pangea' map generates an island somewhere, otherwise I don't care about the shape of the maps or fixing ice caps or making starts fair or anything like that.

 

[intutama]

Some random thoughts...

Does this mean every AI starts with aggr/wheel/hunting in your simulations? It would be interesting to see the statistical changes between "with bonus tech" and "without bonus tech", to see if there are AIs that are penalized by their crappy starting techs. It's not uncommon to see some AIs ignore the wheel for quite a while (I remember a tournament game with Bismarck getting insanely late wheel and aggriculture for example).

It would be interesting if some people good with statistics (i.e. not me) could extract outliers from the numbers and figure out if the data is biased or not, and what factors influence a leader's win rate. For example, coastal vs non-coastal start, or access to poison-stone/marble. But, that woud be quite an undertaking I suppose. Other ideas would be peace-weight impact. A thorough statistical analysis of the game would require quite a lot of of effort both in data collection and data analysis.

Finally, the principal reason to watch the Sullla AI game in my opinion is because the inherent randomness makes it fun - stupid decisions by the AI, random wars across the map, insane runaways, insane comebacks, and so on. I think that's why people like the Wildcard game (it has more views on youtube), because it's even more AIs and raging barbarians. And Sullla is generally entertaining to watch.

 

[Keler]

Does this mean every AI starts with aggr/wheel/hunting in your simulations?

It would be interesting if some people good with statistics (i.e. not me) could extract outliers from the numbers and figure out if the data is biased or not, and what factors influence a leader's win rate. For example, coastal vs non-coastal start, or access to poison-stone/marble. But, that woud be quite an undertaking I suppose. Other ideas would be peace-weight impact. A thorough statistical analysis of the game would require quite a lot of of effort both in data collection and data analysis.

Finally, the principal reason to watch the Sullla AI game in my opinion is because the inherent randomness makes it fun - stupid decisions by the AI, random wars across the map, insane runaways, insane comebacks, and so on. I think that's why people like the Wildcard game (it has more views on youtube), because it's even more AIs and raging barbarians. And Sullla is generally entertaining to watch.

I don't understand if you asked to me or ManiaMuse89

Every AI starts with two techs only in my simulations. Just like in noble difficulty.

Since all games autoplayed just one time, I don't have any outlier data or not. I just give pure randomness a challange here eventually hoping it would detect a better ai from worse after thousands of try.

There are so many factors to effect win rate surely. Aside from AI illustrated, I think the easiest way to find which starting positions are better/worse on a map is putting same leader AI on every starting position and run AHs. Mine is a random attempt to find that out. I did not think of going in details of what kind of starting positions are often stronger on a map but they are there in pictures above. Some maps most dominant start is a coastal capital, such as season 6 pericles start ,other maps there are strong central starts.

I'm sorry. This game has so much mechanics I can't cope with everything. I guess I could
-run same games giving barbs archery techs
-run same games giving ais deity bonus techs
But it is just too much to do for comparing.

 

[Keler]

SEASON 8

Spoiler Map Name: S8G1 :

Spoiler Map Name: S8G2 :

Spoiler Map Name: S8G3 :

Spoiler Map Name: S8G4 :

Spoiler Map Name: S8G5 :

Spoiler Map Name: S8G6 :

Spoiler Map Name: S8G7 :

Spoiler Map Name: S8G8 :

Spoiler Map Name: S8 Wildcard1 :

Spoiler Map Name: S8 Wildcard2 :

Spoiler Map Name: S8PO1 :

Spoiler Map Name: S8PO2 :

Spoiler Map Name: S8PO3 :

Spoiler Map Name: S8 Final :

 

[Thrasybulos]

Playoff3 : seems I was correct to consider SB's starting position superior to Churchill's?

Championship: one of these is not the same.

 

[Keler]

You were correct for Playoff3 based on what I find out

As for championship map, most games ended before turn 300,and they are spaces! That "not same" Gandhi start is probably coincidence unlucky ,technically Victoria should have won too but Hatshepsut stole it. Overall map favors space over culture one sidedly, if by any badluck some cultural ais like Huayna Capac turns on culture slider during superconducts he basically loses it despite having 10 tech lead. Next season if we see the map again, we can blindly click on space victory condition, turn 290.

 

[Keler]

I am posting this to MAYBE criticize AI survivor's tournament format.

I was wondering how much current tournament format favors diplo strenght over anything else of an AI.
As you can see peaceweight 2-3-4 heavily favored and always does better so I calculated how the ranking would be based on only openning round game results. This way ranking is fairer?
Left is main ranking, right is oppening round results only ranking, yeah you can understand from number of appeared columns

2437 game results vs 1452 game results.
A genious would say 186 tournament x 8 is = 1488
However I used the power of flexibilty and made a couple of 6 and 5 oppening round game tournaments to be able to use 9 or 11 player wildcard maps. Same logic continued.
For 5 and 6 oppening round games 10 or 12 leaders made into playoffs. All other survivors go to wildcard. The only rule is playoffs has max capacity of 18 leaders. Other than winner and runner up, rest of best survivors made into playoffs too but they earned no points in the process. If there are 2 wildcard games (more than 11 oppening round game survivors) then third place survivors have privilage regardless of end game score. Rest of survivors sorted through score to fill the remaining spots of playoffs.
So lets say if less than 18 leaders ended up surviving for playoffs. Then only 2 playoffs made where the third survivors (or third and fourth survivor if other game has no survivor) made into final game. If no survivors in playoffs too? Yeah that'd suck but never happened Final gets canceled then. Final game can't be less than 6 leader surviving.
Score system remained always same, winner 5, runner up 2 points.

I hope that explains something!

 

[Thrasybulos]

Yes, owing to the simple fact there are more low peaceweight AIs than high peaceweights, the elimination format of AI survivor does favour low peaceweights in the final rounds (playoffs + championship).

That said... I wouldn't say that the differences between your two lists are that dramatic, actually?

 

[Fippy]

Every time i see those lists it's staggering how we had Stalin & Charlie as champions, while Hatty didn't score a single point