Thanks, RefSteel! It was a fun season. I just wish I knew how I did it, so I could try to do well again in future.
I am curious to see how the seeds change when the results of season 8 are added into the totals. A few leaders will certainly climb the rankings after a very good season, while some with disappointing results will slide. Does anyone know if there were changes to pool one and pool two membership?
(September 29th, 2024, 11:49)haphazard1 Wrote: Thanks, RefSteel! It was a fun season. I just wish I knew how I did it, so I could try to do well again in future.
I am curious to see how the seeds change when the results of season 8 are added into the totals. A few leaders will certainly climb the rankings after a very good season, while some with disappointing results will slide. Does anyone know if there were changes to pool one and pool two membership?
I am not sure, but if I am not mistaken pool 1 remains unchanged, while three leaders drops out pool 2.
In pool 1 Huayna Capac (55 points) is dethroned by Mansa Musa (58 points) who did have good season. Justinian (43 points) is sole leader in 40s.
Suryvaraman and Gilgamesh have equal points (33 points), however due Gilgamesh failing to advance and Suryvaraman going to wildcard, Suryvaraman stays in pool 1 and Gilgamesh in pool 2.
Notably Gandhi did rise in pool 2 and with 32 points he is close to pool 1 membership.
Then in pool 2, Loius (25 points), Hannibal (24 points) and Mao (25 points) drops out and enter Mehmed (28 points), Cyrus (27 points) and Churchill (25 points and reached finals this season)
Elizabeth (7 --> 20), Augustus (10 --> 22), Mansa Musa (48 --> 58) and Churchill (16 --> 25) are biggest climbers this season.
Alhambram is correct, except that he forgot that Gilgamesh placed second to reach the playoffs this season. That puts him back in Pool One and Sury in Pool Two.
IND is probably the worst trait on Deity due to additive bonus but does okay anyway because of good AIs propping it up. ORG still seems useful because Deity AIs can crash their economies just fine. Maybe before BtS before they nerfed the difficulty it wasn't possible for the AIs to crash and ORG would be the worst, but not anymore.
Thanks, Alhambram and Eauxps. Several of the biggest climbers were starting from a very low level, and are still not seeded despite their big gains in season 8. But at least they are not pathetic no-hopers any longer.
(October 1st, 2024, 11:19)MJW (ya that one) Wrote: What are the trait ratings now?
IND is probably the worst trait on Deity due to additive bonus but does okay anyway because of good AIs propping it up. ORG still seems useful because Deity AIs can crash their economies just fine. Maybe before BtS before they nerfed the difficulty it wasn't possible for the AIs to crash and ORG would be the worst, but not anymore.
I've been working on pulling in the Alternate Histories' results into a dataset and analyzing some of these things (through Season 7). This gives us 1,077 games with 7,075 finishes to work with. And "which traits are good" was a large part of why I started this.
Structurally, a question - how important is "surviving a match" - whether it matters if an AI's 3rd and alive vs 3rd and dead? I didn't grab that data point, and while I don't want to go back and re-do that, if we need it we need it.
Also defined "an adjusted finishing place" - treating each game as evenly placing leaders from 1st to 6th no matter the game size. This means 3.5 is always the average adjusted place, so anything below 3.5 is good and anything above is not. In a 7-player game, the adjusted places are 1.00, 1.83, 2.67, 3.5, 4.33, 5.17 and 6.00 place, with FTD always at 6.00; otherwise the 11-player WC games would seriously distort the data, especially since most leaders aren't in that game. And for the alternate histories, we don't always have score for the non-top 2 survivors, so in a 6-player game with 4 surviving players, it'll be 1.00, 2.00, 3.50, 3.50, 5.00 and 6.00 place. Similar logic is used in the rare times where two AIs die on the same turn.
A few things I've found so far:
On traits, Cre and Ind are the most consistent, and Fin is the most likely to get first.
1st place: Fin 22.7%, Ind 20.9%, Cre 20.6% (bottom 2 are Agg at 10.7% and Cha at 10.6%)
2nd Place: Imp 18.7%, Ind 17.3%, Cre 16.9% (Fin is lowest at 10.9%)
FTD: Phi 21.9%, Spi 20.8%, Fin 16.9% (least-likely are Cre at 11.8% and Ind at 11.6%)
Adjusted Place: Cre 3.19, Ind 3.21, Fin 3.42 (Phi is worst at 3.87)
On peaceweight, a peaceweight of 1-4 are all beneficial - the mean adjusted place is 3.11 - 3.22 for each of those, while 0 (mean: 3.60) and the high peaceweights (6, 8, 9) (means 3.75 - 3.83) all underperform, and Gandhi as the lone peaceweight 10 is an especially poor performer (4.38), because he's a FTD outlier (37.3%, with Monty, Asoka, Ragnar, and Lincoln all being FTD 29-32% as the next-most likely).
Justinian really stands out as being very good despite his traits - his traits do not have good results, despite him being one of the strongest leaders in the dataset. In 77 games, Justinian has never been first to die. Note: he's been in the fewest games because his non-AH AI Survivor luck has been pretty bad in Seasons 5-7 when we've consistently done Alternate Histories, and in Season 4 we didn't choose his games (game 4 and Playoff 1) for the Alternate Histories that season.
(October 1st, 2024, 11:19)MJW (ya that one) Wrote: What are the trait ratings now?
IND is probably the worst trait on Deity due to additive bonus but does okay anyway because of good AIs propping it up. ORG still seems useful because Deity AIs can crash their economies just fine. Maybe before BtS before they nerfed the difficulty it wasn't possible for the AIs to crash and ORG would be the worst, but not anymore.
I've been working on pulling in the Alternate Histories' results into a dataset and analyzing some of these things (through Season 7). This gives us 1,077 games with 7,075 finishes to work with. And "which traits are good" was a large part of why I started this.
Structurally, a question - how important is "surviving a match" - whether it matters if an AI's 3rd and alive vs 3rd and dead? I didn't grab that data point, and while I don't want to go back and re-do that, if we need it we need it.
Also defined "an adjusted finishing place" - treating each game as evenly placing leaders from 1st to 6th no matter the game size. This means 3.5 is always the average adjusted place, so anything below 3.5 is good and anything above is not. In a 7-player game, the adjusted places are 1.00, 1.83, 2.67, 3.5, 4.33, 5.17 and 6.00 place, with FTD always at 6.00; otherwise the 11-player WC games would seriously distort the data, especially since most leaders aren't in that game. And for the alternate histories, we don't always have score for the non-top 2 survivors, so in a 6-player game with 4 surviving players, it'll be 1.00, 2.00, 3.50, 3.50, 5.00 and 6.00 place. Similar logic is used in the rare times where two AIs die on the same turn.
A few things I've found so far:
On traits, Cre and Ind are the most consistent, and Fin is the most likely to get first.
1st place: Fin 22.7%, Ind 20.9%, Cre 20.6% (bottom 2 are Agg at 10.7% and Cha at 10.6%)
2nd Place: Imp 18.7%, Ind 17.3%, Cre 16.9% (Fin is lowest at 10.9%)
FTD: Phi 21.9%, Spi 20.8%, Fin 16.9% (least-likely are Cre at 11.8% and Ind at 11.6%)
Adjusted Place: Cre 3.19, Ind 3.21, Fin 3.42 (Phi is worst at 3.87)
On peaceweight, a peaceweight of 1-4 are all beneficial - the mean adjusted place is 3.11 - 3.22 for each of those, while 0 (mean: 3.60) and the high peaceweights (6, 8, 9) (means 3.75 - 3.83) all underperform, and Gandhi as the lone peaceweight 10 is an especially poor performer (4.38), because he's a FTD outlier (37.3%, with Monty, Asoka, Ragnar, and Lincoln all being FTD 29-32% as the next-most likely).
Justinian really stands out as being very good despite his traits - his traits do not have good results, despite him being one of the strongest leaders in the dataset. In 77 games, Justinian has never been first to die. Note: he's been in the fewest games because his non-AH AI Survivor luck has been pretty bad in Seasons 5-7 when we've consistently done Alternate Histories, and in Season 4 we didn't choose his games (game 4 and Playoff 1) for the Alternate Histories that season.
Oh cool, I've been wondering about trying to crunch the numbers on Alternate History data, but never actually did the legwork to aggregate it all. Is your database shareable?
(October 1st, 2024, 11:19)MJW (ya that one) Wrote: What are the trait ratings now?
IND is probably the worst trait on Deity due to additive bonus but does okay anyway because of good AIs propping it up. ORG still seems useful because Deity AIs can crash their economies just fine. Maybe before BtS before they nerfed the difficulty it wasn't possible for the AIs to crash and ORG would be the worst, but not anymore.
I've been working on pulling in the Alternate Histories' results into a dataset and analyzing some of these things (through Season 7). This gives us 1,077 games with 7,075 finishes to work with. And "which traits are good" was a large part of why I started this.
Structurally, a question - how important is "surviving a match" - whether it matters if an AI's 3rd and alive vs 3rd and dead? I didn't grab that data point, and while I don't want to go back and re-do that, if we need it we need it.
Also defined "an adjusted finishing place" - treating each game as evenly placing leaders from 1st to 6th no matter the game size. This means 3.5 is always the average adjusted place, so anything below 3.5 is good and anything above is not. In a 7-player game, the adjusted places are 1.00, 1.83, 2.67, 3.5, 4.33, 5.17 and 6.00 place, with FTD always at 6.00; otherwise the 11-player WC games would seriously distort the data, especially since most leaders aren't in that game. And for the alternate histories, we don't always have score for the non-top 2 survivors, so in a 6-player game with 4 surviving players, it'll be 1.00, 2.00, 3.50, 3.50, 5.00 and 6.00 place. Similar logic is used in the rare times where two AIs die on the same turn.
A few things I've found so far:
On traits, Cre and Ind are the most consistent, and Fin is the most likely to get first.
1st place: Fin 22.7%, Ind 20.9%, Cre 20.6% (bottom 2 are Agg at 10.7% and Cha at 10.6%)
2nd Place: Imp 18.7%, Ind 17.3%, Cre 16.9% (Fin is lowest at 10.9%)
FTD: Phi 21.9%, Spi 20.8%, Fin 16.9% (least-likely are Cre at 11.8% and Ind at 11.6%)
Adjusted Place: Cre 3.19, Ind 3.21, Fin 3.42 (Phi is worst at 3.87)
On peaceweight, a peaceweight of 1-4 are all beneficial - the mean adjusted place is 3.11 - 3.22 for each of those, while 0 (mean: 3.60) and the high peaceweights (6, 8, 9) (means 3.75 - 3.83) all underperform, and Gandhi as the lone peaceweight 10 is an especially poor performer (4.38), because he's a FTD outlier (37.3%, with Monty, Asoka, Ragnar, and Lincoln all being FTD 29-32% as the next-most likely).
Justinian really stands out as being very good despite his traits - his traits do not have good results, despite him being one of the strongest leaders in the dataset. In 77 games, Justinian has never been first to die. Note: he's been in the fewest games because his non-AH AI Survivor luck has been pretty bad in Seasons 5-7 when we've consistently done Alternate Histories, and in Season 4 we didn't choose his games (game 4 and Playoff 1) for the Alternate Histories that season.
Oh cool, I've been wondering about trying to crunch the numbers on Alternate History data, but never actually did the legwork to aggregate it all. Is your database shareable?
Here's the source files (the AI Leader Info is just generic traits/peaceweight). I still need to work on the visualizations/summaries, and probably testing a few regression models. Alternate Count of "0" means "Not alternate history."
(October 16th, 2024, 20:35)Cyneheard Wrote: Here's the source files (the AI Leader Info is just generic traits/peaceweight). I still need to work on the visualizations/summaries, and probably testing a few regression models. Alternate Count of "0" means "Not alternate history."
I just wish I knew how I did it, so I could try to do well again in future. 
Several of the biggest climbers were starting from a very low level, and are still not seeded despite their big gains in season 8. But at least they are not pathetic no-hopers any longer.
