New Hotfix Version (12-15)

[IcyAngel]

What difficulty and handicap numbers are you using? I've quit playing for now because the AI gets either so much culture it's crazy or not enough of other stuff.

We really need AI to get 1/2 culture compared to other yields on handicap bonuses.

We'd need to determine to how much of the total culture pool handicap yields account for first; if handicap culture is 2/3 of the pool and you cut it in half, that means the AI is getting 33% less policies than it did before. I don't know about you, but in my games that would be a complete disaster for AIs.

 

[BiteInTheMark]

I wasn't sure if puppet cities use the same logic, but you are correct, they ignore monuments (and wells/watermillls) for a really long time

Iam 100% sure, they use the same logic as AI. The former puppet-build-logic caused heavy problems (using farming, even a lot of buildings were available). G said, he had changed it to the normal AI behavior.

We'd need to determine to how much of the total culture pool handicap yields account for first; if handicap culture is 2/3 of the pool and you cut it in half, that means the AI is getting 33% less policies than it did before. I don't know about you, but in my games that would be a complete disaster for AIs.

But if we solve such silly problems the AI have with honoring culture buildings, this will maybe compensate the cut in culture.
The ultimative goald would be a competitive AI, which dont need any buffs. (I know, this never will happen, but less necessary artificial yields would make balancing more easier.)

 

[Infixo]

Great solution. The problem is just that if you get 20 cities with banks, you earn too much gold, enough for basically infinite army and it destroys all poverty, eliminating one of the biggest problems of building such a massive empire

Out of curiosity, I’ve checked how income structure looks like in my favorite Inca example. Wide civ, 22 cities. Mid game. Making ~2400 gold, spending around ~1900. 20 banks, 16 caravansaries. Each caravansary producing 26-32 gold, varies between cities. So, on average 16*29=464. Plus city modifiers, varies 10-30% between cities, lets say 15%. It gives a total 534. 20% of entire income from one building.
I vote to nerf this. Either remove, put limit or activate the boost only if there’s a trade route between Caravansary and the Bank.

 

[Infixo]

Its an easy way to test the monument thing. Conquered and puppeted cities have strange build order. Its very annoying, puppet cites prefer building universities, banks and other stuff, but not a 1 or 2 turn monument. Leading to the result, my puppet cities have problems in working tiles, cause they ignore easy buildings, especially culture buildings..

This is actually very easy to explain. Whenever there is something to build, AI calculates value of that, both units and buildings. This algorithm is quite complex, but essentialy it converts all possible gains into „points”, similar to yields. It is 2-pass, then it is sorted and best one is chosen out of top 3-2-1, depending on difficulty.
Important thing is: all calculations take into account the current situation and possible gains based on that. No future is analyzed, no complex connections between buildings, etc. Not possible for that level of AI. So, if you have a monument that gives flat 2 culture vs. Public School that gives many more yields, the choice is obvious for AI.

 

[Blue Ghost]

This is actually very easy to explain. Whenever there is something to build, AI calculates value of that, both units and buildings. This algorithm is quite complex, but essentialy it converts all possible gains into „points”, similar to yields. It is 2-pass, then it is sorted and best one is chosen out of top 3-2-1, depending on difficulty.
Important thing is: all calculations take into account the current situation and possible gains based on that. No future is analyzed, no complex connections between buildings, etc. Not possible for that level of AI. So, if you have a monument that gives flat 2 culture vs. Public School that gives many more yields, the choice is obvious for AI.

Does the AI not take production costs into account? That seems like an obvious oversight.

 

[Infixo]

Does the AI not take production costs into account? That seems like an obvious oversight.

It does, ofc. But it may not be enough for a Monument to outweight later buildings. Cost is translated into turns to build, and 1 is minimum ofc, even if it is theoretically like a fraction of a turn. If something takes 5 turns, it gets 1,5x less points. 10 turns - 2x less. 16 turns - 3x less. It's not linear, ofc.
Point is, if those advanced buildings take 5-10 turns, and they might since AI gets production bonuses, 2x less points is not enough. I suppose advanced buildings are weighted easily 10x-20x more than Monument.
Also: you can check the numbers in CityStrategyAIProductionLog.csv.
And just as an example. Ohrid is a conquered city, a Puppet.

275, Denmark, Ohrid, PRE: Building, Customs House, 27180, 10
275, Denmark, Ohrid, PRE: Building, Forge, 11610, 3
275, Denmark, Ohrid, PRE: Building, Harbor, 9550, 10
275, Denmark, Ohrid, PRE: Building, Windmill, 9030, 20
275, Denmark, Ohrid, PRE: Building, Well, 7745, 1
275, Denmark, Ohrid, PRE: Building, Council, 6685, 1
275, Denmark, Ohrid, PRE: Building, Herbalist, 6580, 2
<MANY OTHER LINES>
275, Denmark, Ohrid, PRE: Building, Monument, 320, 1
275, Denmark, Ohrid, PRE: Building, Constabulary, 300, 16
<2ND PASS, MANY OTHER LINES>
275, Denmark, Ohrid, SEED: 10016956444603654428, CHOSEN: Forge, Do not rush, TURNS: 3

Forge got more points in 2nd pass than Customs House, Monument didn't make into 2nd pass.

 

[tu_79]

It does, ofc. But it may not be enough for a Monument to outweight later buildings. Cost is translated into turns to build, and 1 is minimum ofc, even if it is theoretically like a fraction of a turn. If something takes 5 turns, it gets 1,5x less points. 10 turns - 2x less. 16 turns - 3x less. It's not linear, ofc.
Point is, if those advanced buildings take 5-10 turns, and they might since AI gets production bonuses, 2x less points is not enough. I suppose advanced buildings are weighted easily 10x-20x more than Monument.
Also: you can check the numbers in CityStrategyAIProductionLog.csv.
And just as an example. Ohrid is a conquered city, a Puppet.

Forge got more points in 2nd pass than Customs House, Monument didn't make into 2nd pass.

Probably building Customs House first is the best option, IF the city already has some sort of culture. Does it take into account unhappiness? I mean, if a city is suffering from boredom, does it increase the value of culture?

 

[Infixo]

Probably building Customs House first is the best option, IF the city already has some sort of culture. Does it take into account unhappiness? I mean, if a city is suffering from boredom, does it increase the value of culture?

Yes, it accounts for Happiness, too. Buildings that counter a type of unhapiness that exists in that city are given more points. Also, buildings' Flavors are accounted for along the way (e.g. Monument is only 10 flavor culture, Customs has 100 gold, 40 culture). So, all adds up and results you can see in the log.

 

[BiteInTheMark]

It does, ofc. But it may not be enough for a Monument to outweight later buildings. Cost is translated into turns to build, and 1 is minimum ofc, even if it is theoretically like a fraction of a turn. If something takes 5 turns, it gets 1,5x less points. 10 turns - 2x less. 16 turns - 3x less. It's not linear, ofc.
Point is, if those advanced buildings take 5-10 turns, and they might since AI gets production bonuses, 2x less points is not enough. I suppose advanced buildings are weighted easily 10x-20x more than Monument.
Also: you can check the numbers in CityStrategyAIProductionLog.csv.
And just as an example. Ohrid is a conquered city, a Puppet.

Forge got more points in 2nd pass than Customs House, Monument didn't make into 2nd pass.

This is very interesting. Was this decision mechanic ever changed/touched by the VP mod?
Does the Ai also include, that it requires some buildings as a prerequisite? Or are locked buildings invisible to them?
I think, if we can fix this decision making for a more forward planning AI, this will lead to much much stronger AI.

 

[Deadstarre]

Cost is translated into turns to build, and 1 is minimum ofc, even if it is theoretically like a fraction of a turn.

that is really unfortunate for them (and necessary by coding im sure), I am regularly taking into account the production surplus when building.

However I think even in that case if monument could have been weighted into a fraction, in your example it still wouldn't have made the second pass, they just seem to be undervaluing it.

 

[Infixo]

This is very interesting. Was this decision mechanic ever changed/touched by the VP mod?

Yes, there are many changes due to VP, all new features are programmed.
I also found that it boosts buildings from earlier eras, based on era difference. So, it seems that everything is programmed correctly, maybe just weights need to be adjusted.

 

[Blue Ghost]

Why are production costs not weighed linearly? It seems to me that a building with half the build time is exactly twice as good to build.

 

[Infixo]

Why are production costs not weighed linearly? It seems to me that a building with half the build time is exactly twice as good to build.

Civ5 uses exponential formulas like everywhere...

 

[tu_79]

Civ5 uses exponential formulas like everywhere...

Shouldn't cheaper buildings get more priority? For Progress, it's an added bonus. For the rest, the sooner a building is operating, the sooner it gives yields. As you explained it, it sounds like more expensive buildings get to be built before cheaper ones.

 

[Ziad]

it sounds like more expensive buildings get to be built before cheaper ones.

If this is true, then couldn't this lead to "starvation" where the civ is stuck queuing high cost buildings as they unlock newer and more expensive buildings?

Might be why some civs never get proper infrastructure rolling.

 

[hokath]

If you watch puppets you have in the late game this is what happens
edit: I say late game, any time you have cities that are undeveloped relative to tech

 

[Infixo]

Shouldn't cheaper buildings get more priority? For Progress, it's an added bonus. For the rest, the sooner a building is operating, the sooner it gives yields. As you explained it, it sounds like more expensive buildings get to be built before cheaper ones.

Assuming that 2 buildings are valued the same (let's say, 1000 points), the one that takes LESS turns will have priority. Everything's ok here. Only its turn-weighted value is not linear, it uses expotential formula for a divisor actually.
Edit. But if entire algorithm is taken into consideration and we're discussing things that a city can build, since they are valued very differently, it happens very often that something that takes more turns to build takes priority.

 

[Deadstarre]

But if entire algorithm is taken into consideration and we're discussing things that a city can build, since they are valued very differently, it happens very often that something that takes more turns to build takes priority.

there is only a handful of buildings I feel that way about, like no matter how long theyre going to take I need to get on them ASAP. Trainstation is one of them (most of them have to do with hammers) but its usually a complex thought process. as a general rule, building the cheapest buildings first is one that would most often mimick my usual game behavior (not that im an authority on building stuff, but i feel like i am!)

 

[CrazyG]

It does, ofc. But it may not be enough for a Monument to outweight later buildings.

I think my point can be best summarized as. "The Monument should outweigh later buildings if the AI is playing well"

There are 0 circumstances where a wall should outweigh a monument. Looking at that list, the monument and well should almost certainly be #1. This is a specific example which appears to come up quite often (my puppets do this too, are puppets and AI using the same code?). Based on your explanation, it sounds like production cost needs more consideration, it shouldn't be just a tie breaker.

 

[tu_79]

Assuming that 2 buildings are valued the same (let's say, 1000 points), the one that takes LESS turns will have priority. Everything's ok here. Only its turn-weighted value is not linear, it uses expotential formula for a divisor actually.
Edit. But if entire algorithm is taken into consideration and we're discussing things that a city can build, since they are valued very differently, it happens very often that something that takes more turns to build takes priority.

Still. It sounds like it should be weighted the other way.