Last Pre-Integration AI Games Analysis (4.21.1)

[L. Vern]

Hello everyone,
ran some AI game automation on the last pre-integration patch - zero reproducible CTDs and remarkably well balanced, hope you find the stats interesting!
No 4UC this time, just base VP on 8 player Emperor Communitu games.

Spoiler Victory Types :

Spoiler Authority Victories :

Authority victories are a fun thought experiment to see how well warmongers actually do - basically, if at any point during the game a civ controls at least 67% of plots and 67% of cities, it is retroactively assigned a victory. These charts explore what the distribution of wins looks like under authority victory rules as well as the the actual outcome of games that ended in an authority victory:

Spoiler Civ Winrates :

tiles and cities columns refer to average percentage ownership per game

edit: the forum seems to have lost the ability to render markdown tables correctly. I'll attach the table if/when I figure it out

Spoiler Technology Research Times :

Spoiler Era Progression Times and Attainment :

Spoiler Policy Adoption and Winrates :

Spoiler Yield Sources by Era :

Cities
Instant Yields
Handicap Yields
Misc.

 

[L. Vern]

Spoiler Largest Instant/Handicap Yield Sources Breakdown :

Spoiler Religion Attainment Times :

Spoiler Pantheons :

Spoiler Founders Beliefs :

Spoiler Enhancer Beliefs + Follower Beliefs Chosen at Enhance Time :

Spoiler Reformation Beliefs :

 

[Alpakinator]

Thank you for doing these analyses. Good to see, victory types are more balanced.

Why are there 3 different sets of data for border instant yields across ages in Instant Yields Largest Contributor Triggers? One in shades of green, one in brown and one in orange? Is it for food, production gold? then blue for science and purple for culture?

What does n(avg) mean? Does it mean, across all games each AI on average got only 3.9912 instant yield triggers from border growth that gave production in modern era? That seems way too low, given that AI on average got 2720 of that yield (production?) form border growth.

 

[andersw]

So all top are AIs are strong CS contenders and the AI's are notoriously bad at shutting down CS play ie just conquering the CS to prevent stuff like this.
A few surprises regarding top civs, Babylon would definitely be much higher if you asked me to guess beforehand and Poland does much worse in my games.
I've experienced India AI bugged (not literally but yes I have disabled india in my games), their bonus growth stuff +mandirs mean they go bananas over the top broken op so not surprised to see India as highest non CS civ (at 4th).
Religious options match pretty much the common order AI takes stuff in, I see even more open sky and mandir is 100% taken 1st, faith of the masses go first maybe 95%.
I wonder what celts would look like if you forced them to take morrigan (its sooo op).

 

[L. Vern]

Thank you for doing these analyses. Good to see, victory types are more balanced.

Why are there 3 different sets of data for border instant yields across ages in Instant Yields Largest Contributor Triggers? One in shades of green, one in brown and one in orange? Is it for food, production gold? then blue for science and purple for culture?

What does n(avg) mean? Does it mean, across all games each AI on average got only 3.9912 instant yield triggers from border growth that gave production in modern era? That seems way too low, given that AI on average got 2720 of that yield (production?) form border growth.

Yes, the colors correspond to yields - and they're also labelled on the right of the heatmaps, though I suppose it's fairly easy to miss

n_avg is more difficult to interpret - it takes into account all civ-game instances, not only civs that made it to modern and had at least one instance of that trigger. So because most civs never make it to modern, being treated as 0 they bring the average down significantly.

I think you're right that an average of 681 per trigger is a bit high - I think I may have a bug in the aggregation code that adds up all of the yields from a trigger per turn and treats that as a single instance of the bonus yields being granted. I'll look into it and get back to you on that

 

[hokath]

Tutelary Gods is a big outlier!

+1 Faith, +3 Production, and +2 Gold if the City has at least 3 Citizens. +1 Faith and Production from Engineers.

I wonder if the reason is that it encourages the AI to work Engineers, and they otherwise are undervaluing them?

 

[axatin]

Tutelary Gods is a big outlier!

I wonder if the reason is that it encourages the AI to work Engineers, and they otherwise are undervaluing them?

That might be the case, the city governor does take into account pantheon yields, so the AI should work more Engineers if they take Tutelary Gods. But I think it's also important that they get production and gold almost unconditionally in every city, which helps them gain an early advantage. Also, it's a pantheon that's easy to use for the AI, they don't need to change their style of play to get the bonuses.

It would be interesting to see the religion found rates by pantheon, I assume they are able to found consistently with Tutelary Gods, but not with pantheons like God of War.

 

[andersw]

That might be the case, the city governor does take into account pantheon yields, so the AI should work more Engineers if they take Tutelary Gods. But I think it's also important that they get production and gold almost unconditionally in every city, which helps them gain an early advantage. Also, it's a pantheon that's easy to use for the AI, they don't need to change their style of play to get the bonuses.

It would be interesting to see the religion found rates by pantheon, I assume they are able to found consistently with Tutelary Gods, but not with pantheons like God of War.

Not so sure, tutelary is a low faith yield pantheon and I would say they mostly found because a. they were 1st or 2nd with pantheon or b. they were aztec/spain.
Ie just that fact that they pic a pantheon early helps to found.

 

[L. Vern]

That might be the case, the city governor does take into account pantheon yields, so the AI should work more Engineers if they take Tutelary Gods. But I think it's also important that they get production and gold almost unconditionally in every city, which helps them gain an early advantage. Also, it's a pantheon that's easy to use for the AI, they don't need to change their style of play to get the bonuses.

It would be interesting to see the religion found rates by pantheon, I assume they are able to found consistently with Tutelary Gods, but not with pantheons like God of War.

Not the cleanest graph I've made but hopefully should get the point across. I was a bit surprised at how much worse some picks like Goddess of Festivals are than similarly picked pantheons like Ancestor worship. The tutelary gods numbers were also much lower than I would have thought, though 75% is nothing to scoff at of course

 

[azum4roll]

AI is so bad at founding nowadays in general. Refusing to build shrines is a huge problem.

Goddess of Festivals is known to be bad. Hopefully we can rebalance it better with decimal yields later.

 

[hokath]

@L. Vern Would it be possible to see a visualization of number of policies against time?
I'd like to fix up the Enlightenment Era mod and it seems to me like the main challenge is altering Policy costs to scale appropriately, especially so as not to upset the Victory Time Spread at the top.

 

[L. Vern]

@L. Vern Would it be possible to see a visualization of number of policies against time?
I'd like to fix up the Enlightenment Era mod and it seems to me like the main challenge is altering Policy costs to scale appropriately, especially so as not to upset the Victory Time Spread at the top.

Sure, here's a vis that shows the timing of policy acquisitions

Annotation boxes refer to mean turn a policy tree from that era is acquired.

At the moment I don't have the data for every policy picked within a tree, only when the tree is acquired. If the timing of each individual policy within a tree is what you're after I can look into if that's available in the logs

 

[azum4roll]

That seems to imply some players got the first policy on turn 0?

 

[msw1]

Why does the AI love mandirs so much? It's always the top recommended belief for me too. It seems like junk to me and when I've taken that belief I've never noticed it doing anything other than producing a trivial amount of yields. But maybe I don't understand it.

 

[L. Vern]

That seems to imply some players got the first policy on turn 0?

No, just that this type of graph (kdeplot) assumes a gaussian distribution, and I chose the number of underlying allowed parameters to be pretty small to get a smoother looking line. If we look at at more allowed paramaters and the observations overlayed to get a better idea

then yea it's clear it's a falsehood that's a result of the under-parametrization. I only really wanted to get general trends across but I suppose it's more accurate to show it with more detail

 

[azum4roll]

Who got Artistry before turn 80?

 

[hokath]

Probably Poland?

I love overlaying the raw data on the histograms, it's my favorite thing.
Artistry being slightly earlier makes sense. I suppose same logic explains why Imperialism is often later -- civs in this position probably need to pay more attention to catching up, perhaps Industry was the better choice for some of these guys?

If the timing of each individual policy within a tree is what you're after I can look into if that's available in the logs

What I'd like to do is adjust the Policy Cost formula of

cost = 50 + (4x+0.2)**2.22

(whoever commented this code didn't know what exponential scaling was but ok)
to take into account the addition of N new techs.

So if I had the turn each policy was adopted I could back out a total culture at turn n and from that interpolate an effective culture rate.
Then use the technology research times figure to give me the number of turns, T, I expect to be added by the N new techs.
Then adjust the formula accordingly so the same (or, if balancing, maybe fewer) policies are adopted in the additional T turns; subject to changing the acquisition before Renaissance as little as possible (on your figure, roughly turn 180).

Does that sound right to you?

 

[L. Vern]

Yep, Poland consistently picks the fastest