A Better AI.

[Solver]

I'm running a test to see how the starting positions are affected. I'm not going to play full games with this right now, but I start several new games and examine my surroundings.

The games are started under Fractal, which is the map I usually play on.

1st start:



Coastal start, no seafood in the immediate vicinity. Only Corn and Horses in the capital's fat X, of the early resources. Some Calendar resources nearby.

2nd start:



Only Pigs and Gold in the immediate vicinity. Some Floodplains, but not enough to go crazy. Other than Gold, no happy resources too close to me, not a great start either.

3rd start:



1 source of clams! But that's it, fish outside of the city's range. However, Clams + Corn + Pigs are good food for the city. Not bad, but looking at the surrounding terrain, doesn't seem great, either.

4th start:



No seafood! No immediately available happy resources nearby. A spot of desert that is disappointing. Two, even. Again, not the best start.

5th start:



Two Clams, finally! Still, no other resources in the capital's fat X. A disappointing desert, but nice founding spots along the river.

6th start:


The biggest feature of this start is along the river. You can found a city with access to 4 Calendar resources, for uber-commerce. The capital seems better than the previous starts. Food AND happy.


My overall impression is, these starts are a bit too balanced. I understand the changes affect the area immediately around the capital. Which seems, in all starts, average, a bit better or a bit worse. Never really bad, never great. I would've expected to draw at least one extremely good start in 6.

Comments?

[Blake]

Remove the DLL and generate 6 more starts.

edit: I generated about 4 starts and look what I got on the 4th try...


[img]http://img183.imageshack.us/img183/5245/holy****gj6.jpg[/img]


2 Clams, Grassland Gem, Grassland Hill Gold, Freshwater Corn plus all 3 important early strategics in the outer fat cross!
The disturbing thing is my changes would not have effected that start in any way - it would've turned out exactly that way in stock 2.08.


By the way I'll outline exactly what it'll do different now:
1) It'll choose the site on the map differently, favoring higher land site and less bad land.
2) INSIDE the fat cross the resources generated might differ slightly.
3) Both inside and outside the fat cross for a single tile radius (and only a single tile) it will upgrade the TERRAIN TYPE, it originally did this only in the fat cross itself. There's only about a 50% chance of it actually upgrading outside the fat cross, unless it's coastal.
4) It will remove bad features in an additional 1 tile belt around the fat cross, at a 50% chance on land. For a 2 tile belt at sea, with a 100% chance on coast and a 50% chance in deeper water.


RESOURCES OUTSIDE THE FAT CROSS WILL NOT CHANGE. Only the terrain type and bad feature removal very near the fat cross. In other words if you see change further away you are imaging things. AT NO POINT ARE RESOURCES OUTSIDE THE FAT CROSS EVER ACCOUNTED FOR IN START POINT EVALUATION. NOT NOW, AND NOT BEFORE EITHER. Only the bare terrain yield was accounted for (and only barely), now the "shape" of the land is also a factor, too much water and it's penalized slightly.

[Nacht]

Originally posted by Blake
Ugh how is it that a non-null tile within a city radius isn't getting worked by the city? I'd like to know, mostly out of sheer morbid curiosity. Got the save or know the circumstances?

I think you should test the cultural ownership of the plots. AI cities that want to steal other civs plots because they think they are more important are not nice.

[Solver]

1st start:



May be just me, but I really like it. Many floodplain tiles + Rice in the starting area provides a nice food farm!

2nd start:



Poor. 1 square in-land, average terrain, no seafood anyway. Of course, the Settler should be moved 1 tile south before settling.

3rd start:



Low food until the resources get improved. And no happy resources anywhere nearby that are soon available.

4th start:


Start on Rice, also Fish and 2 Sugar within the fat X. Grassland tiles, so I think it's very solid, given that the capital is coastal.

5th start:



Very solid capital. That's 2 seafood + Pigs in the fat X, plus Iron soon after that. The Jungle in the north can be a bit annoying, but not big enough to be a real problem.

6th start:


4 floodplains, Pigs, Corn and Horses within capital's fat X. Nice spot on the other end of the river, too.


From this, it indeed seems to me that the stock 2.08 starts are more diverse and less balanced, whereas your modded starts are somewhat normalized to be like around-average.

[Solver]

Blake,

What does "INSIDE the fat cross the resources generated might differ slightly" mean? How and why do they differ? Do they differ because of different terrain types, or have you specifically changed the probability of resources being in that fat X?

And what do you call bad features? I assume, stuff like Ice and Jungle?

That's one awesome start you posted there, though .

[Guynemer]

Thanks Blake!

[Blake]

Um I'll check exactly what I changed.

In terms of resource generation, firstly the value is "pulled up" a bit less, meaning the best starts (the one other starts are measured by) will be the same, while poorer starts will be upgraded less. I'm going to adjust this again for consistency but it's not really a big deal.

The second change is that it's less likely to generate seafood. If the city is "Water Biased" (more than half water) then it's only half as likely to generate a water bonus, since it only places a limited number of bonuses this means it's more likely to place a land bonus.

Next it's hardcoded to not add more than 2 coastal resources and more than 2 ocean resources - you could get 2 clams and 2 fish, but you wont get 4 clams.

So pretty much the only change is it's more likely to add land food rather than sea food, especially at sites with excessive sea.


That's pretty much it. You wont sea (see ) any change at all in the resources of a landlocked city, on average anyway. Well you might see changes but they aren't really there.

[Solver]

Ah, so you have changed the seafood allocation explicitly. Is it one of those things that is required for the AI's correct evaluation? Interesting if so. Dunno, I've simply always enjoyed the occasions when a city pops up with 4-5 seafood. Happens rarely enough, but is fun nonetheless .

[Blake]

"Perception":

Only Pigs and Gold in the immediate vicinity. Some Floodplains, but not enough to go crazy. Other than Gold, no happy resources too close to me, not a great start either.

Start on Rice, also Fish and 2 Sugar within the fat X. Grassland tiles, so I think it's very solid, given that the capital is coastal.

I think that speaks for itself. How fish beats river pigs, rice under city beats river gold, 2 sugars beat 5 floodplain and coast beats a large river.
"Not a Great Start"
"Very Solid Start"

[Solver]

You don't agree that seafood resources are better than land food?

[Veritass]

Does the new AI require more memory?

I have a large map that I am in the Renaissance on, and I am slowdowns and starting to get hangs or crashes every few turns. I am very happy with how this game is progressing, and I would hate to just dump it and start again.

I have played large maps before without problems. Does the new AI require more memory than before? Is there any way the new AI would affect video card usage (the weak link in my system)?

I will switch off the new AI and try to continue and see if that makes a difference. Thanks.

[Blake]

It depends on the start. A start with a strong hammer production (plains hill forest is good, or two plains forest) is good for cranking out a workboat. A site with very poor hammers (like your fish/sugar one) is horrible for getting the workboat out. If you can get a workboat out faster than a worker then I tend to think they are great, otherwise not so great.

I'd definitely rather pop a worker out to hook up river pigs, than a workboat for ocean fish, there's no contest. The worker goes on to do more improvements.

Fish are good - 5f tiles always are. 6f tiles are better.

edit:

Is it one of those things that is required for the AI's correct evaluation? Interesting if so. Dunno, I've simply always enjoyed the occasions when a city pops up with 4-5 seafood. Happens rarely enough, but is fun nonetheless .

To answer this, no. It's just something that is inconsistent - you don't get 4 rice in a row, but you would get 4 clams in a row.
There is still the possibility of getting 2 clams and 2 fish, so a 4 seafood start is possible.

generally I think those starts are cruel, after producing 4 workboats that's 120h - that's as much as 2 Workers or Stonehenge. 2 workers would do a lot more work than improving 4 tiles. 1 fish is nice, 2 seafood's okay, more than that gets expensive.
The 5 seafood starts do mess with the AI... they can end up with size 20 cities by 1AD and stuff... the city isn't necessarily that productive because it's extremely unhealthy... the AI doesn't really manage such sites well, and they are very extreme. You wouldn't real mistake it for being balanced.

[Solver]

Well, granted, you're a better player than me and understand this better. I've always liked seafood a lot because I can build a workboat, and my city is growing while I do so. Plus, less research required. Even if you don't start with Fishing, it's a very cheap tech.

[Willem]

Originally posted by joncnunn
2.
On the imporoved AI city placement, I really like it, but the one major thing it's missing is considering if the new location is close enough to the rest of their empire to be defended. See the latest AU game where the Cyprus AI in many(most?) of them pouched an excelent city site too far from their empire (and too close to the humans) to defend, promptly resulting in the human taking the city from them. Maybe the alogrithm could go something like (on standard map; scale for different sizes) if another player's culture boundary is within 6 tiles of proposed city site and on this landmass and I don't have culture already on this landmass within 6 tiles of this proposed city site, then don't build here.

I agree. I've noticed that the AI will leave large gaps inbetween many of it's cities, and leave his empire scattered and spread out. It's just an invitation for me to pick off the ones on the outskirts, or to build cities inbetween and possibly take them through culture. If the AI is going to go out and build a city for a resource, it should fill in the gaps later and deveop a more cohesive empire.

3.
And on the AI bonuses & penalties. Any section where the AI gets bonsuses on Noble is an indicator of where the vanilla unmodified AI is weak in a specific area. The stronger the bonus, the weaker the AI was. So anything that improves the AI in that regard so lower the AI bonsues in proportion to how much the AI was improved.

IMO, Noble doesn't need any AI bonuses at all anymore. I've eliminated them all in my game, at least the ones I can access, and it's doing quite well for itself now. It should serve as the basis for the game, with higher levels giving the AI advantages, and the lower ones giving them to the human.

[Willem]

Originally posted by Solver
From this, it indeed seems to me that the stock 2.08 starts are more diverse and less balanced, whereas your modded starts are somewhat normalized to be like around-average.

Looking at those last starts, of the 6, only one of them could be considered an inland start, and even that was fairly close to the coast. So my observation of starting mostly on or near the coast in default Civ stands. If Blake can give us a bit more variety that way, I'm all for it.