I wrote a paper on a 650 AI Game data analysis (VP 4.5)

[L. Vern]

Some of you may be familiar with the AI autoplay game analyses I have been doing in this forum in the past. This semester, I expanded the scope of this initiative to a graduate data science course project where I explored additional topics and provided a written analysis that I hope you will find interesting!

As it is already formatted in a PDF, I hope you will forgive the lack of a formatted forum post as I have customarily done in the past. The preamble on pages 1-3 can safely be skipped by members of this community.

I would be honored if you gave it a read through and would love to hear your thoughts on the analysis, whether you agree, disagree, have ideas for further exploration that may be visited in future posts or just want to comment or add to something

 

[Phosphoraptor]

1 thing that stands out is the part about the disproportionate GP handicap yields, meaning that civs in AI games that are stronger due to GP focus may not be so strong for a player. This needs to be considered more for balance proposals regarding GP focused civs.

 

[N.Core]

1 thing that stands out is the part about the disproportionate GP handicap yields, meaning that civs in AI games that are stronger due to GP focus may not be so strong for a player. This needs to be considered more for balance proposals regarding GP focused civs.

Also, another thing to note is that Austria's GP modifier was applied twice. That's why Austria was always the leader in the chart. This bug has been fixed and now waiting to be released as an update. This will nerf Austria a lot.

 

[hokath]

From an academic presentation standpoint, I would encourage more complete figure captions that tell the reader more directly what they are meant to see (even if it repeats some things in the text).
It's a general truism that students undercook the captions.
E.g. in Figure 2 you could say what the vertical green line means. You might also tell the reader how many civs are statistically above average, average, and below average and what we're meant to think about that (grammar: "on winrate" in first sentence is redundant)
Figure 12 is better. Avoid being overly colloquial with "fun fact" etc., usually doesn't come off well in marking

I'll read over the actual content later. Looks great. I was amazed religion founding time had median=mean. Do you have an explanation for that?

 

[L. Vern]

From an academic presentation standpoint, I would encourage more complete figure captions that tell the reader more directly what they are meant to see (even if it repeats some things in the text).
It's a general truism that students undercook the captions.
E.g. in Figure 2 you could say what the vertical green line means. You might also tell the reader how many civs are statistically above average, average, and below average and what we're meant to think about that (grammar: "on winrate" in first sentence is redundant)
Figure 12 is better. Avoid being overly colloquial with "fun fact" etc., usually doesn't come off well in marking

I'll read over the actual content later. Looks great. I was amazed religion founding time had median=mean. Do you have an explanation for that?

Thanks, I appreciate your feedback and suggestions! My PhD colleague had many of the same criticisms when reviewing an earlier draft regarding figure captions lol - I added a bit of info after that but clearly still an insufficient amount, good to know. I like your suggestions for additional clarifying information, this is definitely something I'll keep in mind for future technical writing!
Regarding colloquialisms: In this case, as its more of an informal report rather than something being submitted for publication I decided to keep the fun facts as the instructions for the project were verbatim "do cool data science" and my prof that will be marking this likes fun facts LOL. For similar reasons, a lot of wording is a bit clunky and "unscientific", I am aware that words like "certainly" and "clearly" usually don't have any place in an academic publication of this nature but rewriting a lot of sections to sound nicer was just not something I prioritized particularly highly ¯\_(ツ)_/¯

 

[N.Core]

For similar reasons, a lot of wording is a bit clunky and "unscientific", I am aware that words like "certainly" and "clearly" usually don't have any place in an academic publication of this nature but rewriting a lot of sections to sound nicer was just not something I prioritized particularly highly ¯\_(ツ)_/¯

Yeah, it's better that you don't affirm the data with words like "certainly", "definitely", or "clearly". Let the data itself affirm the analysis and conclusion.

 

[azum4roll]

Domination Victory in VP is defined differently. You and your vassals need to own all original capitals to win.

 

[Chippy Smith]

do you have policy pick rate by civ in the logs?

 

[L. Vern]

do you have policy pick rate by civ in the logs?

I finally got around to formatting the table and remembered you were looking for this, here ya go

Spoiler Policy Pick Count by Civ :

 

[azum4roll]

Do you see any significant difference of policy choices in the 4UC data?

 

[rkkn]

lots of 100% authority, a few 100% tradition, and zero 100% progress (though Assyria comes close at 95%)

and why the heck do the shoshone have a 100% artistry pick rate?

 

[Chippy Smith]

Thanks! 4UC data would be interesting too.

 

[pineappledan]

Yeah something’s going desperately off for policy choice decisions. Lots of examples of AI civs that 100% miss some of the most obvious synergies for their kits with these policy choices

 

[L. Vern]

Do you see any significant difference of policy choices in the 4UC data?

No

 

[axatin]

Policy choices are heavily influenced by player traits. Based on the traits, civs are categorized as "Smaller", "Expansionist", "Warmonger", "Religious", "Diplomat", "Tourism" and "Science", and the categories are used in the weighting of the policy choices. Other factors that affect policy choices are the values in the database table "Leader_Flavors" and the victory type the civ is going for.

Spoiler :

It seems to be a bug that all civs are "Religious", I'll check it.

 

[Stalker0]

My look at the policy stats.

Ancient Era

• Arabia should never take anything other than tradition.
• Authority Brazil seems weird, but its not completely out there.
• Tradition Germany seems weird to me long term.
• No Authority for Greece is strange.
• No Authority for Japan is very very weird.
• Authority Spain is off to me, she's not a warmonger.
• Netherlands as tradition seems weird.
• Shoshone as authority is off.

Medieval

• Assyria as heavy fealty is off.
• Austria should just be hard-coded statecraft
• Brazil should be strong artistry, CV is just such a strong part of their kit
• China, its interesting that they are always artistry. not that I think its a bad play, but China has such a flexible style I'm surprised its not more varied here.
• Germany taking artistry, forget about it.
• India taking Artistry might at first glance seem a bit nuts instead of fealty for their religion play....but I do think India is a strong CV play right now with all the holy sites, so it might not be crazy.
• Russia as fealty....I guess its for the border growth speed, ok I don't hate that.
• Inca as pure fealty feels odd.
• Shoshone as artistry feels strange, yeah their stuff is all over the place.

Industrial

• America is imperialism, I get that they have set up Washington as a super warmonger so pushing that play makes sense, but Industry brings a lot of sauce to America's table imo
• Japan not going imperialism is a crime unless they have TTGOG set up. you can literally faith buy your way into super GPs!
• Persia imperialism isn't the worst, but I think there is a lot of rationalism synergy their too for this to be so one sided.
• Portugal very surprised not to see more imperialism considering their water focus. I don't use it all the time as Portugal but a good portion at least.

Ideology

• China, surprised they are so focused on freedom. I think freedom is fine, but I would expect to see a lot of order in their as well.
• Portugal not going Autocracy -> Lebensaum hurts my soul.
• Its generally interesting that the pure warmongers really debate between Autocracy and Order. They were like 100% pure on the others, but when it gets to ideologies they start to waver.

 

[rkkn]

It seems to be a bug that all civs are "Religious", I'll check it.

how about Korea being both "smaller" and "expansionist"
I'm also surprised that Maya don't have the science trait

• Russia as fealty....I guess its for the border growth speed, ok I don't hate that.
• Inca as pure fealty feels odd.

what would you prefer?
I see fealty as the generic tree. if you aren't focused on great works or CS allies, you take fealty

 

[pineappledan]

Polynesia picking artistry only 6% of the time despite culture being their primary win condition.

Many civs that are major warmongers are picking fealty way more than they should. There is no way Huns is founding 100% of the time, so them picking fealty just helps someone else spread. Meanwhile major religious civs like Ethiopia never pick it.

 

[Chippy Smith]

Fixing some of the stranger patterns by adjusting leader flavors seems like a roundabout way of addressing the problem.Why not have policy selection based on victory type first, then civ traits, then proximity and identity/threat of neighbors?

 

[L. Vern]

and why the heck do the shoshone have a 100% artistry pick rate?

Shoshones picks was also something that stood out like a sore thumb the first time I looked at the table, quite unexpected. According to the matrix Axatin posted they have the tourism trait so that might have something to do with it? Still quite strange...

Many civs that are major warmongers are picking fealty way more than they should

Similarly, Axatin mentioned that everyone has the religious trait so that might contribute to this. I also agree with the assessment I saw earlier that it's the closest thing to a generic tree that non culture non diplo civs pick when both of the other options are meh.

Ideology

• China, surprised they are so focused on freedom. I think freedom is fine, but I would expect to see a lot of order in their as well.
• Portugal not going Autocracy -> Lebensaum hurts my soul.
• Its generally interesting that the pure warmongers really debate between Autocracy and Order. They were like 100% pure on the others, but when it gets to ideologies they start to waver.

When it comes to Ideologies there's also a lot more compounding factors like vassalage and unhappiness from non-majority world ideologies forcing them to switch, so it doesn't surprise me that there's no 100% pickrates like we saw in policies. Overall though loved your post, I think you did a great summary that brings the major issues to attention. Certainly looks like there's a ton of room for improvement. The Spain policy picks make 100% sense to me though given what we know about her leader flavors LOL