Terrain city bonus - Page 2

MrBaggins · February 8, 2003, 15:24

well... it is... but its a lot of work:

you create separate visibility AND fog of war maps for each AI civ... and calculate them each turn/as new land is discovered... then only apply knowledge to those squares that are visible. In certain cases, land would need be 'anticipated'. VERY difficult.

Too much work.

NOT cheating is:

* slower
* much more complex
* did I mention slower?

Why not just give it these smarts and give it a penalty?

The thought of needing to give the AI a penalty, so its fair to the human is quite amusing... don't you think?

Gilgamensch · February 8, 2003, 15:37

Wasn't sure about..........

No you didn't mention slower, you REALLY think it is slower

It is amusing

But it always depends on the programmer writing the code

MrBaggins · February 8, 2003, 15:47

True... although I should probably mention, in my mod... since "colonization" (settling without settlers) happens for everyone, even the human, that the cheating isn't really cheating, since everyone gets this advantage.

However, I am clear that tactically... the AI could be made smart by things like knowing when and where armies on transports are coming, and how threatening they are... Yes... it cheats visibility... but the AI's ability to wage war, and defend, is SO poor... that in my mind, its entirely fair.

Imagine moving in on a city, with a large stack, and finding that the AI had reinforced it, and rushed a city walls... before you got to attack...

The only way you can make the AI a challenge is to simulate intelligence. The only way you can simulate intelligence (more) completely is to include prediction. The only way you can simulate prediction with the limited resources that we have, is to have prior knowledge.

AI's knowing that you are building up troops.. and beginning to move them towards you, and countering by also building up theirs is just a wonderful vision, in my mind.

If, ultimately, you need to be very skillful and cunning, to beat the AI... and need to use diplomacy, and so on... all the better.

It just improves the game.

MrBaggins

Martin the Dane · February 9, 2003, 07:48

Just a hort comment from a human player

The first part of this thread, the ICS part, I think is overrecting. When I play I seldom have the option of filling out the gov.-city-limit due to capitol distance and to little land to occupy. I could alternatively take over all my neighbours, but that wouldn't be so much fun.

As a matter of fact I think the value of having more cities is quite balanced. If there is no incentive to build more than one city, tere wouldn't be much of a game. After-all building Settlers take a lot of effort in the early game.

A much better way of balancing the value of large vs. many cities would be to allow ecsess production to be carried over to the next turn. I have often thought that having a city capable of producing 2000 production points only being able to produce one unit/turn even if the unit costs 1000 production, is quite stupid.

Or when building something and missing it by 1 production, the city wastes 1999 production to complete it. This is comparable to a builder building a house and missing one dor knob going out to orders a truckload of materials, just to dump all except the dorknob at the nearest dumpsite.

Now to the discussion of making the AI "smarter": well there is only one thing to say, by all means teach it to select land and defend it a little better. I have seen an AI player with 10+ cities more than half of those within one square of the coast but not one single costal city. Even though some of the cities were far to close to one another.

mapfi · February 9, 2003, 09:04

Quote:

would be to allow ecsess production to be carried over to the next turn

if that could be done that'd be great

Maquiladora · February 9, 2003, 09:25

Yes pleeeeease make the AI place its cities better. At the moment it selects it on the overall score for the terrain but doesnt take enough consideration into where its other cities are. Yeah theres minimum settle distance or whatever but thats too general. It needs to be told more specifically where surrounding borders are, this would not only make the cities grow better later but it would save space for it to build more cities. If that makes it boring and repetitive to look at, i dont care. It would make the AI a better late game challenge.

In a way, REX (rapid early expansion) is more suitable to describe how CtP2 works. You build settlers wherever and whenever you can until you reach the city limit. ICS is like packing cities in 1 tile away from each other infinitely.

I really dont mind the AI cheating, as long as its in a smart way, and not just giving it huge research or growth advantages. Thats more like playing a scenario than a full game.

MrBaggins · February 9, 2003, 10:34

Quote:

Originally posted by Martin the Dane
Just a hort comment from a human player

The first part of this thread, the ICS part, I think is overrecting. When I play I seldom have the option of filling out the gov.-city-limit due to capitol distance and to little land to occupy. I could alternatively take over all my neighbours, but that wouldn't be so much fun.

As a matter of fact I think the value of having more cities is quite balanced. If there is no incentive to build more than one city, tere wouldn't be much of a game. After-all building Settlers take a lot of effort in the early game.

At Impossible, isolated from a neighbor... I can outdevelop their economy and hence win, by the Middle Ages/Rennaisance or the late Rennaisance at the absolute worse.

The priority is always more cities, most science and smart pw. Your general science aim is government oriented. I find such expansion is generally slowed relatively little by the government advances.

With neighbors its a slightly different game, but mainly uses the developmental strength and relatively weak tactical nature of the AI against it.

There is no reason that countries with relatively little land to build cities should be crippled by their inability to spread out. In the modern world, especially, its clear that territory size bears little resemblance to the relative power, except in the grossest case.

Nations such as Japan are absolute anomalies.

Quote:

A much better way of balancing the value of large vs. many cities would be to allow ecsess production to be carried over to the next turn. I have often thought that having a city capable of producing 2000 production points only being able to produce one unit/turn even if the unit costs 1000 production, is quite stupid.

Or when building something and missing it by 1 production, the city wastes 1999 production to complete it. This is comparable to a builder building a house and missing one dor knob going out to orders a truckload of materials, just to dump all except the dorknob at the nearest dumpsite.

It *MAY* be possible to do so... maybe.

A more important change would be to create 'OCC' style governments, that rather improving in bureaucratic ability to govern larger empires at a distance, concentrated on their ability to govern and make efficient very small nations... like those with an empire cap of 3. As the governments advance, the cap never increases, yet the coefs increase significantly, especially in Gold and Science.

You'd have a separate path of government, and would need to force an government that started this way to not expand by massively penalizing such... like -10 happiness penalties. You'd need to ensure that the AI doesn't ignore the cap (it does now: AI civs regularly expand beyond their cap, small map or not.) Government change could effectively be only one way; you could change to an expansive government from an isolationist government, but not the other way.

Quote:

Now to the discussion of making the AI "smarter": well there is only one thing to say, by all means teach it to select land and defend it a little better. I have seen an AI player with 10+ cities more than half of those within one square of the coast but not one single costal city. Even though some of the cities were far to close to one another. *SNIP*

True... the settling behavior of the AI should be taken out of its hands *COMPLETELY*.

MrBaggins

MrBaggins · February 9, 2003, 10:54

Quote:

Originally posted by Maquiladora
Yes pleeeeease make the AI place its cities better. At the moment it selects it on the overall score for the terrain but doesnt take enough consideration into where its other cities are. Yeah theres minimum settle distance or whatever but thats too general. It needs to be told more specifically where surrounding borders are, this would not only make the cities grow better later but it would save space for it to build more cities. If that makes it boring and repetitive to look at, i dont care. It would make the AI a better late game challenge.

There are a number of factors which the AI doesn't take into effect, in city placement, such as continent size/shape, and presence and proximity of other civs.

Add to this that the AI fudges the MinCityDistance when it feels like it... and you have a problem...

'Borders' aren't something that are easily or quickly discovered and correlated. Comparitive city distances are, and they achieve the same thing. The most useful thing to know is how close each other city on the same continent, is to this one...

And I do agree with you... maintaining the challenge is what its all about.

Quote:

In a way, REX (rapid early expansion) is more suitable to describe how CtP2 works. You build settlers wherever and whenever you can until you reach the city limit. ICS is like packing cities in 1 tile away from each other infinitely. *SNIP*

A pretty good attempt to curtail ICS was made, and many of the components were put in place. ICS is always there, its just a little tougher to achieve.

I agree that you could call this something different again... but that would ultimately, just be a semantical game.

The final key of the puzzle, is making the production and commerce more closely match the proportion of population.

MrBaggins

Maquiladora · February 9, 2003, 11:48

So basically the aim is to make it possible to "out-research" a bigger empire, if you specifically aim to do this, obviously sacrificing military and/or production. Of course youd still need a fairly big empire to begin with, but outgrowing and "out-tileimping" them only, needs to be changed.

At the moment its better to stay in Communism (after Fascism) than switch to Democracy, even for the science gains, because you generate more PW under Communism, you can upgrade all your commerce tile imps incredibly quickly, and overtake a Democracy (amount of cities on each side being equal) and you can still support a massive military, which you cant under Democracy.

Also there should be bigger benefits for having contact with more Nations, and having good relations, reguarly exchanging/buying techs etc. An isolated Nation from the dawn of civilization to modern day would find itself way behind in technology when it met the first new civ, simple as that. Yeah it will grow as quickly as any other nation, but it should be held back by the isolation, like it should need to build a granary to grow above size 4 or something, but it cant until it researches agriculture, for example. Went off topic abit.

Gilgamensch · February 10, 2003, 08:22

Quote:

Originally posted by MrBaggins
True... although I should probably mention, in my mod... since "colonization" (settling without settlers) happens for everyone, even the human, that the cheating isn't really cheating, since everyone gets this advantage.

Forgotten this one, so no cheating and equal situation

Quote:

However, I am clear that tactically... the AI could be made smart by things like knowing when and where armies on transports are coming, and how threatening they are... Yes... it cheats visibility... but the AI's ability to wage war, and defend, is SO poor... that in my mind, its entirely fair.

Imagine moving in on a city, with a large stack, and finding that the AI had reinforced it, and rushed a city walls... before you got to attack...

Depends, imagine you come with a armie of 6/6 (Attacker/Ranged), and you knew he only had like 3 defenders, because he took the city just before, and suddenly you find city walls + Tower + 6/6 stack defending it, woudl be quite a surprise.....

But general sounds like a

A little bit more information, like seeing transport(s) coming and therefore strengthening the city, is a different thing. Not sure how easy this is..........

Quote:

The only way you can make the AI a challenge is to simulate intelligence. The only way you can simulate intelligence (more) completely is to include prediction. The only way you can simulate prediction with the limited resources that we have, is to have prior knowledge.

AI's knowing that you are building up troops.. and beginning to move them towards you, and countering by also building up theirs is just a wonderful vision, in my mind.

If, ultimately, you need to be very skillful and cunning, to beat the AI... and need to use diplomacy, and so on... all the better.

It just improves the game.

MrBaggins

Agreed, but I just want to prevent that you'll need another computer to beat it...........

MrBaggins · February 10, 2003, 11:01

Quote:

Originally posted by Gilgamensch
*SNIP*
A little bit more information, like seeing transport(s) coming and therefore strengthening the city, is a different thing. Not sure how easy this is..........
*SNIP*

Here's how the 'seeing something coming' technique works:

A transport starts a long way from this opponent. It is a long distance from all of its cities, but proportionately, they are all about the same...

The closer it gets to a particular map point, the proportionately lower the distance is from the transport to the city it will land near.

You strengthen cities defences based on 'certainty': you concentrate your defence in the cities closer to the landing point/ approach point.

MrBaggins

Gilgamensch · February 10, 2003, 11:42

Quote:

Originally posted by MrBaggins

Here's how the 'seeing something coming' technique works:

A transport starts a long way from this opponent. It is a long distance from all of its cities, but proportionately, they are all about the same...

The closer it gets to a particular map point, the proportionately lower the distance is from the transport to the city it will land near.

You strengthen cities defences based on 'certainty': you concentrate your defence in the cities closer to the landing point/ approach point.

MrBaggins

OK, now another possible exploit, or a possible crash:

You send 2 transports equally equipped, what to calculate now?
Or the exploit:
You send like three transports, 1 with a 12 army-stack, and the other two with 2 * 6 stacks.
The computer would 'only consider the first one a threat (or main threat) and with the rest combined, you would have an attack -force.

Don't want to sound to pessimistic, just some points to think about.

This problem would be even more problem if you have only like roads or even railroad, movement is too restricted. The human could go for the edges and a minimum in the middle and could take those, the AI would move to the middle, leaving the edges undefended............

MrBaggins · February 10, 2003, 11:59

You include threat ratings based on force composition of the respective armies.

All inverse proportions*threat are totalled... giving a 'picture' of how to defend. More incoming threats will indeed, even out the overall picture, but this can be countered by making bigger threats more exponentially than smoothly rated. IE... a 12 stack is 'worth' far more than a 6 stack.

The biggest threat receive the most attention.

Its basically a *LOT* more sophisticated than what is happening now.

MrBaggins

Gilgamensch · February 10, 2003, 12:27

Quote:

The biggest threat receive the most attention.

But again, if you use one 12 stack as a bait and use as attack-force like 3(6) * 4 stacks, the AI would see the bigger one as the most dangerous one, 'forgetting' the option that the human could join those little stacks to one or two big ones. Just with roads and a good planning (from the human) side and the AI would be doomed.

I hope you get what I mean

MrBaggins · February 10, 2003, 12:56

Solved by intoducing the idea of threat groups... grouping threats having similar (adjustable) distance proportions.

Since its a system that objectifies uncertainty, it will never be precise...

BUT

any threat management system will be an improvement over the current "even-dispersal and hope for the best" system thats in place now.

Its clear that the AI certainly can out-produce a human... in a recent game, I attacked an AI opponent on a medium sized island, where he had in excess of 500 units to my 350. I was attacking with a force of 50 units... If he force matched (and put a higher priority on bombard troops and techniques) and counter attacked- expecting losses, but knowing that he had many many more to throw at the 'problem'... I would not have been able to perform the same techniques... I would have needed many, many more troops to do the same task... a massive undertaking...

MrBaggins

Gilgamensch · February 11, 2003, 04:18

How do you want to make the AI recognising groups?

Also if all units are the same distance how should the AI evaluate?

Let's make it easy, you attack with 36 warriors,
1 army (12 units), 2 armies with 6 each, and the rest in groups of 2.

And they are overall at the same distance. For the

human, he could easily re-group the groups and attack with any stack. But how does the AI want to defend against it?

Again let's make it easy, the AI would have 10 cities, 2 in the middle and the rest on a circle around it. The human would attack with the above mentioned stacks from all directions, will the AI also use it's 'internal' defenders to distribute around, trying to equalize the forces?

Mentioning the 'bad' behaviour: Is it somewhere a place in SLIC, where you determine, which stacks the AI should use? Could it be done, depending on the oponents abilities?

All this would kind of include some cheating, but .........

Examples:

The enemy has produced bombers, therefore assign fighters as garrision.
The enemy is using stacks of tanks, so produce bomber-stacks to attack.
The enemy has canon (whatever long range), so include those as well into stacks.

Reason, why I mention it, I had it, we all had artillery and he was still using his old bombard (not sure about the name, playing the German version by now......). Not really useful.......

About the numbers and his behaviour:

In another thread something was mentioned about two values in one of the files, which gave 'priority' to either pillaging or conquering. Don't know if you read this one........

MrBaggins · February 11, 2003, 08:58

Pictographic example.

Sea scale (to the left) is condensed.

Numbers on the island are 'free defenders'.

MrBaggins · February 11, 2003, 09:01

example 2... notice the new disposition of the free defenders.

MrBaggins · February 11, 2003, 09:08

example 3... notice that the attacked side is further favored, moreso where the larger threat is present.

MrBaggins · February 11, 2003, 09:16

Groupings are done as follows:

The distance from each each unit 'threat' is measured from each of your cities.

This generates a 'virtual direction'.

Other units which are close to each other will have similar virtual directions... that is the percentage of distances that closely match will be large.

Units that match to a certain percentage are thus grouped, for purposes of threat determination.

Gilgamensch · February 11, 2003, 09:19

For attacks coming from the same site, it would work (quite nicely), but what if you would have stacks coming from all directions? (espacially if you would have 3*6 attacking in your case from the east?)

MrBaggins · February 11, 2003, 09:31

Quote:

Originally posted by Gilgamensch
*SNIP*
And they are overall at the same distance. For the

human, he could easily re-group the groups and attack with any stack. But how does the AI want to defend against it?

'Regrouping' will ultimately mean that the grouped units are ultimately closer to one point, than another.

Initially, the AI will have a generalized sense of defensive concentration. The closer the threat gets, the more definite it is.

Quote:

Again let's make it easy, the AI would have 10 cities, 2 in the middle and the rest on a circle around it. The human would attack with the above mentioned stacks from all directions, will the AI also use it's 'internal' defenders to distribute around, trying to equalize the forces?

Eventually, yes... the closer the threat the more exponentially threatening it is.

Quote:

Mentioning the 'bad' behaviour: Is it somewhere a place in SLIC, where you determine, which stacks the AI should use? Could it be done, depending on the oponents abilities?

You can't describe the map, or individual threats to the strategic AI... You'd have to override the AI, using SLIC to a certain degree to achieve this.

Quote:

All this would kind of include some cheating, but .........

Examples:

The enemy has produced bombers, therefore assign fighters as garrision.
The enemy is using stacks of tanks, so produce bomber-stacks to attack.
The enemy has canon (whatever long range), so include those as well into stacks.

Reason, why I mention it, I had it, we all had artillery and he was still using his old bombard (not sure about the name, playing the German version by now......). Not really useful.......

Easy to do... you can easily tell what is being produced by an enemy... so you could play a little 'paper-scissors-stone' game... matching the 'threat' with an appropriate defense.

The AI should be well supplied with appropriate tech: it gets automatic upgrades, and coupled with some knowledge dissemination, it should never be so far behind as to be antiquated.

Matching threat with specific defense does complicate things a little (how do you split up troops when you have 'imperfect' resources?)

Quote:

About the numbers and his behaviour:

In another thread something was mentioned about two values in one of the files, which gave 'priority' to either pillaging or conquering. Don't know if you read this one........

Yes, I've read it... Its a little off-topic, but these do need to be at least looked at.

MrBaggins · February 11, 2003, 09:37

Quote:

Originally posted by Gilgamensch
For attacks coming from the same site, it would work (quite nicely), but what if you would have stacks coming from all directions? (espacially if you would have 3*6 attacking in your case from the east?)

If the threats proximities are the west and the east... then the defensive priorities are the west and the east. Closer proximity means a more exponentially 'valuable' threat... so both sides would be more 'threatened' than the center.

In your example, roughly 48 troops per coast would be matching up with the 18 attackers on each coast.

Not bad odds...

MrBaggins

Tamerlin · February 11, 2003, 09:44

All the ideas exposed here are very interesting but I think the AI could be much more threatening if it could be taught to attack when it has an advantage. In my current SAP2 game my main opponent had three stacks of 10 to 12 units while my supply lines were thin. The AI could have launched a counterattack that would have inflicted me several bad blows. Instead of attacking these stacks have been used to pillage my TI which let me the time to strengthen my cities and finally attack the lingering stacks. Frankly, I don't know if I would have been able to recover should the AI had attacked when it was time to do so.

mapfi · February 11, 2003, 09:52

Hey MrBaggins, this concept looks quite interesting. The whole thing should be worked out with Frenzy, producing a real complete overhaul of that script.

on a side note, I don't think we should give the AI knowledge of the enemies troups though.

Gilgamensch · February 11, 2003, 09:58

Quote:

Originally posted by Tamerlin
All the ideas exposed here are very interesting but I think the AI could be much more threatening if it could be taught to attack when it has an advantage. In my current SAP2 game my main opponent had three stacks of 10 to 12 units while my supply lines were thin. The AI could have launched a counterattack that would have inflicted me several bad blows. Instead of attacking these stacks have been used to pillage my TI which let me the time to strengthen my cities and finally attack the lingering stacks. Frankly, I don't know if I would have been able to recover should the AI had attacked when it was time to do so.

Tamerlin,

for your problem try to find the threat, which I mentioned above (sorry, can't remember the title). those values actually change a lot the AI's behaviour. Those values determines weather the AI will attack or just pillage. It was quite recent, so try a search with pillaging.

MrBaggins · February 11, 2003, 09:59

I believe the issue with the non-aggressiveness lies in two areas:

1) There are many more 'targets' of opportunity to pillage or 'HARASS'.

2) The AI is unaware of ablative attacks...
I.E. a single massed assault may not work, but multiple attacks could 'solve' the problem. All 'Free-Attackers' should be organized for efficiency, and then, if viable, attack.

Tamerlin · February 11, 2003, 10:08

Quote:

Originally posted by mapfi
Hey MrBaggins, this concept looks quite interesting. The whole thing should be worked out with Frenzy, producing a real complete overhaul of that script.

The Frenzy code has a disadvantage as diplomacy becomes almost useless. It is extremely difficult to reach an agreement with the AI civs and they don't bother to respect it.

Quote:

...on a side note, I don't think we should give the AI knowledge of the enemies troups though.

I agree with you, this is typically the kind of cheat that can detract me from a game because it considerably reduces the strategical depht of a game (IMHO the Civ3 AI behavior is ridiculous). How can you build a strategy when you don't know the rules your opponent must respect? Though this is certainly coming from the fact I am playing tabletop wargames and boardgames I really don't like this.

Tamerlin · February 11, 2003, 10:13

Quote:

Originally posted by Gilgamensch
Tamerlin,

for your problem try to find the threat, which I mentioned above (sorry, can't remember the title). those values actually change a lot the AI's behaviour. Those values determines weather the AI will attack or just pillage. It was quite recent, so try a search with pillaging.

I have already altered the datas following Dale's advices (I suppose this is the thread you are talking about though I can't remember the name), the AI is more prone to attack with my settings but it can still be improved. The problem is I don't like to tinker with the settings by myself as I don't really know what I am doing. I am new to modding and I don't have any knowledge of programming languages.

Tamerlin · February 11, 2003, 10:17

Quote:

Originally posted by MrBaggins
2) The AI is unaware of ablative attacks...
I.E. a single massed assault may not work, but multiple attacks could 'solve' the problem. All 'Free-Attackers' should be organized for efficiency, and then, if viable, attack.

This is one of the main problem, the AI would have taken over several of my cities without any problem if it had been able to use such a tactic.