AU mod: How to help the AI with research choices - Page 2

Nor Me · December 12, 2003, 14:41

Quote:

Originally posted by alexman
There is a new tech property: Non-tradable.
Perhaps that's the reason? I hope so.

I know the techs were tradeable because the AI traded them to me. In one case, I couldn't trade away the tech I had traded for in the same turn to a civ that I'd checked didn't have it to trade before.

donZappo · December 12, 2003, 18:31

I think that something from this discussion should be added into the initial AU Mod which appears to be set for release in the next week or so. While there are a lot of changes that need extensive testing, I think that a test regarding Fascism/Communism could be quite helpful for testing purposes. It would be easy by using flavors to adjust the value to be equal to, slightly less than, or significantly less than the corresponding techs at the time.

I'm pretty sure that this is the one point that there is the most agreement on, and a simple enough fix to incorporate into the mod. The rest of the major points can wait for a more in-depth analysis.

nbarclay · December 12, 2003, 19:03

We need to be very careful about adding to the priority of Philosophy or Scientific Method in an effort to make the AIs more inclined to race for them. On one hand, such changes would make the player less likely to win the race. But on the other, AIs that prioritize them over techs with a higher intrinsic value and don't benefit from winning the race to them would get hurt. I'm not claiming to have an answer to how best to address this issue, but it's one that needs to be considered.

Nathan

alexman · December 19, 2003, 12:45

After lots of testing, I have come up with a specific proposal for the ancient age.

As mentioned in the first posts, the idea is to make the AI less predictable and more likely to take advantage of its traits.

The following 7 flavors are added: Generic, High, Low, Religious, Scientific, Militaristic, Seafaring.

All civ flavors have a 70% relationship with a the tech flavor 'high'.
All civ flavors have a 0% relationship with a the tech flavor 'low'.
All civ flavors have a 50% relationship with a the tech flavor 'generic'.
A civ with the Generic flavor has a 50% relationship with all flavors except 'high', and 'low'.
A civ with the Religious/Scientific/Militaristic/Seafaring trait has a 80% relationship with itself, and a 50% relationship with all other traits except 'high', and 'low'.

The high and low flavors are not symmetric with respect to 50% so that we can get a larger spectrum of flavors by averaging.

Proposed Technology Flavors:
Bronze Working: Generic, Low
Masonry: Generic
Alphabet: Generic, High, Low, Scientific, Seafaring, Naval Transport
Pottery: Seafaring
The Wheel: Generic, Militaristic
Warrior Code: Generic, Low, Militaristic
Ceremonial Burial: Religious
Iron Working: Generic, Militaristic
Writing: High, Scientific, Seafaring
Mysticism: Generic, Religious
Mathematics: Generic, High
Philosophy: Generic
Code of Laws: Generic
Literature: High, Scientific, Diplomats, Military Alliances
Map Making: Low, Seafaring
Horseback Riding: High, Militaristic
Polytheism: Generic, High, Religious
Currency: Generic, High
The Republic: Generic, High
Monarchy: Generic, High, Religious
Construction: Generic, High, Low

In order to see the effect of the above changes, I ran tests, and the results can be seen in the attached table. I took five snapshots of the tech tree: No techs, only bronze working, all first level techs, all second level techs, and all but the last tech in each branch. The research choices for an AI of each flavor are shown in terms of percentages. Each percentage in the table is based on only 150 tests, so it is not exact. It gives a good idea of the real percentage though.

Dominae · December 19, 2003, 13:52

Woah! I'll have to sit down with a big cup of tea later on tonight to look over all that.

I'll preemptively say

though.

Dominae

ducki · December 19, 2003, 15:23

Quote:

Originally posted by Nor Me

I know the techs were tradeable because the AI traded them to me. In one case, I couldn't trade away the tech I had traded for in the same turn to a civ that I'd checked didn't have it to trade before.

Is it possible you traded FOR it on the AI's turn, then couldn't trade it on your turn due to the AI that traded it TO you also traded it to other AIs?

Just a thought, but I generally kick myself when I make a trade and then realize I did it on the AIs turn, ruining my chances of profitting from it.

Theseus · December 19, 2003, 16:00

Where is REL for Bronze Working (known) though?

Tea? I ain't gonna try to understand this with TEA!! (Wrong Flavor!

) I need something stronger.

Nor Me · December 19, 2003, 17:44

I looked at a quick test. I gave Masonry and the Iroquois Flavor1 and they would pay at most 303 gold for it. When I changed the relationshiop of flavor1 to itself from 100% to 50%, they would pay only 93 gold.

That doesn't look good

.

Does anyone have a sensible explanation as to why I've never seen a protestant France in Age of Discovery?

Theseus · December 19, 2003, 18:12

It looks to me like Seafaring is not appropriately researching Mapmaking in the final group... huh? Also, it looks to me like you may have nerfed it for the non-Seafaring civs by setting it at Low.

But overall: Less predictable? WOW!!

Quote:

The Republic: Generic, High
Monarchy: Generic, High, Religious

Friggin' genius.

vmxa1 · December 19, 2003, 19:28

I would give a provisional thumbs up as well. Check that I revoke the provisional and just flat give you one. That was a lot of effort, well done.

donZappo · December 19, 2003, 23:06

Good job, Alexman. I was trying to figure out a good relationship between tech costs and research percentage but the results I were getting were somewhat mathematically erratic. I was hoping to figure that out to avoid the lengthy process that you must have gone through to run these tests! Thanks for the work on this as it appears quite invaluable! Your results look quite good and I look forward to incorporating them and see how they run myself.

lockstep · December 20, 2003, 07:12

Quote:

Originally posted by Theseus
It looks to me like Seafaring is not appropriately researching Mapmaking in the final group... huh?

As I see it, the final group just means that if a Seafaring civ can choose between a) Mapmaking b) getting out of Despotism, it will rather choose b) and research a government tech - not too bad, IMO. And looking at the 'second level tech' group, it seems that Seafaring civs won't have to choose between Mapmaking and a government tech very often, because they will normally beeline to Mapmaking.

EDIT: BTW, the same effect occurs with Scientific civs and Literature.

alexman:

alexman · December 21, 2003, 07:58

Quote:

Originally posted by Theseus
Where is REL for Bronze Working (known) though?

You can assume it's the same as generic. Religious civs do have Burial as a priority, but since they start with it, it doesn't matter.

Quote:

Originally posted by Nor Me
I looked at a quick test. I gave Masonry and the Iroquois Flavor1 and they would pay at most 303 gold for it. When I changed the relationshiop of flavor1 to itself from 100% to 50%, they would pay only 93 gold.

I did not investigate the effect of flavors on AI trade value, but something like this would obviously be enough to kill the flavor idea.

lockstep · December 21, 2003, 09:21

Quote:

Originally posted by alexman
I did not investigate the effect of flavors on AI trade value, but something like this would obviously be enough to kill the flavor idea.

I'd rather say 'rebalance' than 'kill'. Obviously, if the AI really wants a tech (100% flavor) that another AI already has, it is willing to pay about 3 times as much compared to the tech being 'generic' (50% flavor). Your initial proposal uses figures up to 80% (which should result in a willingness to pay about 2 times as high), and most figures fall within the 40% - 70% range. If the AI really is willing to sacrifice too much gold with your proposal, tune your flavor figures towards 50%. This still should produce more variation and less predictability in the AI's research choices, although not as markedly as in your initial proposal.

alexman · December 21, 2003, 13:30

The problem is that with flavors at around 50%, the AI thinks the techs are worth much less than their actual value. That's fine as long as an AI civ trades tech-for-tech with other AI civs, but it's exploitable in trades with the human player, who would just wait for the AI to do the research and then get them for peanuts by trading.

From a quick test, I see that the trade value of a tech with no flavor is the same as when the tech has somewhere between a 99% and 100% flavor relationship (closer to 99%).

So to preserve balance for tech trading with the AI, the generic flavor would probably have to be at around 99%, the high flavor at 100%, and the low flavor something slightly lower than 99%.

Unfortunately, I can't do any extensive testing before two weeks, to see if that's enough to make a difference in AI research choices (I'm away for the holidays).

alexman · December 30, 2003, 15:20

I did some simple tests to get a feel for the effect of flavors on the AI trade value.

A tech that is traded to the AI for 241g without flavors, is traded for the following amounts with flavors:

Code:

Flavor  Gold
 100%    334
  99%    238
  98%    234
  97%    230
  50%    102
  40%     85
  30%     73
  29%     72
  25%     67

So here are some observations from my tests:

99% is very close to the trade value without flavors.
When there are multiple civ/tech relationships in effect, the trade value depends on the maximum of those relationships.
A tech cannot be traded when either civ has a maximum flavor relationship to the tech of under 25%.

Given these observations, I think it's not a good idea to give a tech a low or high priority to all civs using flavors, but it is still possible to use flavors to take advantage of civ traits.

Giving a Warrior Code a low priority, for example, would reduce its trade value in the eyes of the AI, but the AI would still pursue that tech often because it allows archers. Humans would then just wait to get that tech from the AI for cheap.

But giving Alphabet a high priority (100%) for scientific civs, for example, I think would be fine. The scientific AI would then pay big bucks (well, 40% more gold) to get Alphabet, but that money would be well spent, I think, as it would bring them closer to libraries. From another test I did, the 100% flavor relationship is enough to make a big difference in AI research preferences, compared to no flavor, or a 99% flavor relationship.

The incentive for all AI civs to research a given tech more often would have to be done as we did with the PTW version of the mod, or by similar means, such as adding resources with zero appearance ratio.

Unless someone wants to give it a shot before me, I will try to come up with and test a new proposal regarding ancient age AI research priorities next week when I return from vacation. It will probably involve giving Religious civs a 100% relationship to Mysticism, Polytheism, and Monarchy, Scientific civs a 100% relationship to Alphabet, Writing, and Literature, Seafaring civs a 100% relationship to Writing, Pottery, and Map Making, Militaristic civs a 100% relationship with Iron Working, the Wheel, Horseback Riding, and every flavor a 99% relationship to every other flavor except itself. I'm also thinking we should increase the research value of currency, literature, and horseback riding for all civs by adding three new never-appearing resources that have those techs as a prerequisite.

Theseus · December 30, 2003, 21:17

alexman, I am a little confused... how does this interact with your earlier suggestions?

This *strengthens* the previous results?

If that is the case, I am not sure we should go so far yet. Let's try one or the other approach first, see how it works, and then tweak some more.

My thinking is that we don;t want to completely lock AI civs into research patterns, for two reasons:

* Randomness of environement and thus flexibility is required.

* Prediction, boredom, and possible exploits on the part of players.

alexman · December 31, 2003, 11:03

I agree that we want more unpredictable AI research choices, but this is not possible to do with flavors.

So without any testing, I'm not sure if the latest proposal (well, it's not a proposal yet) strengthens or weakens the previous results, but it's the only option if we want to use flavors and keep the AI trade value for trait non-beeline techs more or less what it is in stock C3C.

Buckets · December 31, 2003, 17:17

It is possible if you're willing to use the 'not tradeable' flag. I am doing this for my own purposes, but i wouldn't be surprised if you found it too radical for AU.

(Imagine..... the ai beelining for replaceable parts.... researching medicine and sanitation.... steampower then industrialization.... a world without hate... but only sometimes...)

alexman · January 6, 2004, 16:33

I did some more testing and I have come up with the following proposal for the ancient age, which affects the AI trade value of each tech as little as possible.

The following 5 flavors are added: Generic, Religious, Scientific, Militaristic, Seafaring.

All civ flavors have a 99% relationship with a the tech flavor 'generic'. This ensures that the AI trade value for a generic technology remains almost the same as with stock C3C.
A civ with the Generic flavor has a 99% relationship with all technology flavors.
A civ with the Religious/Scientific/Militaristic/Seafaring trait has a 100% relationship with itself, and a 99% relationship with all other traits. A side effect of this is that Religious civs will offer about 40% more gold for a Religious technology, a Scientific civ 40% more gold for a Scientific tech, and so on.
A technology with a 99% flavor relationship has a AI higher research priority than if it had no flavor, even though the AI trade value is almost the same as if it had no flavor. This is why some techs have the generic flavor and some don't have any flavor at all.
Dummy resources with a zero appearance ratio are added to give a higher research priority when needed.

Proposed changes to Ancient Age Technologies:
Masonry: Generic
Alphabet: Scientific, Naval Transport
Pottery: Seafaring
The Wheel: Militaristic
Ceremonial Burial: Religious
Writing: Scientific, Seafaring
Mathematics: Generic
Philosophy: Generic, Optional

, Dummy resource
Code of Laws: Generic, Military Alliances, Dummy resource
Literature: Scientific, Military Alliances, Dummy resource
Horseback Riding: Militaristic
Polytheism: Religious, Dummy resource
Currency: Generic, 2 Dummy resources
The Republic: Generic, Military Alliances
Monarchy: Religious
Construction: Generic

I didn't give Map Making a Seafaring flavor because this technology already has a high AI priority and we want to keep some unpredictability. Similarly for Iron Working and the Militaristic trait.

In order to see the effect of the above changes, I ran tests, and the results can be seen in the attached table. I took four snapshots of the tech tree: Bronze working known, all first level techs known, all second level techs known, and all but the last tech in each branch known. The research choices for an AI of each flavor are shown in terms of percentages. Each percentage in the table is based on only 150 tests, so it is not exact. It gives a good idea of the real percentage though.

alexman · January 7, 2004, 16:20

So what do you think? Should I continue for the rest of the ages? There is quite a bit of testing involved, so I'd rather not do it if something along those lines is not going to make it in the mod.

Perhaps we can mark the ancient age proposal as under consideration, and if voted in, we can see how it plays out in the first AU game before continuing for the rest of the ages.

By the way, I suspect that a 100% flavor relationship, as proposed above, will help influence the AI with build preferences. Not sure how to test that though.

Stuie · January 7, 2004, 17:18

It would probably be a good idea to try out the Ancient age as proposed prior to putting in the effort for the other ages. That way if we hate it, you won't have wasted your time.

As for the changes, they look interesting, but I'm having a hard time conceptualizing how they will actually impact a game. I guess I'd like to try a game with the changes and see how it plays.

alexman · January 12, 2004, 17:27

Well, I had some time on my hands, so I finished the proposal for AI build priorities for the rest of the ages.

As previously mentioned, the goal is to make AI research selection more unpredictable, but at the same time more likely to benefit from their traits. The value that an AI civilization offers for a technology should be approximately the same as with stock Civ3. An exception is that AI civs are willing to offer about 40% more gold for techs that match their traits, and similarly charge 40% more gold to trade a tech to a civ with traits that match the technology flavor.

To maintain unpredictability, I tried to avoid giving flavors to heavily favored AI technologies. For example, Astronomy was not given the seafaring flavor, and Invention was not given the Militaristic flavor, even though these technologies would clearly benefit those specific traits.

In order to balance the AI research values, it was necessary to add to some technologies extra properties and a number of dummy (banana) strategic resources, which have an occurrence ratio of zero. These do not influence map generation or game play in any way.

So here is the proposal, which I would like to place under consideration (only changes to technologies are listed):

alexman · January 12, 2004, 17:29

It was not possible to balance the high AI value of Feudalism, Invention, Gunpowder, and Astronomy. The AI will still research these technologies as soon as it is able to do so. In order to balance the rest of the technologies, I took four slices of the tech tree, one of which is not a real slice, because it includes technologies that are prerequisites of each other. I did this just to compare the AI research values, even though it’s not possible for the AI to choose among those techs.

The Militaristic trait is encouraged to take the Cavalry beeline.

The seafaring trait is encouraged to research Navigation, as this technology will enable them to sell their no-doubt significant contacts and become rich. This trait is also encouraged to research Physics when given the chance, so that they get to Magnetism earlier. Magnetism and Astronomy are not given the seafaring trait, because these technologies already have a high value for all the AI civilizations.

Scientific civilizations are encouraged to beeline to Education to get Universities.

Religious civilizations are encouraged to get Monotheism for early Cathedrals.

alexman · January 12, 2004, 17:30

Most technologies in this era were given a higher value, so that Communism and Fascism are much less commonly researched by the AI. Nationalism and Replaceable Parts have such a high AI value that no attempt was made to discourage the AI from researching them as soon as possible.

The seafaring trait is encouraged to sometimes research ironclads, which most other traits ignore. It is also encouraged to research Combustion, which allows many new ships. Mass production also allows ships, but it was not given the seafaring flavor as it has a high value for all traits.

The militaristic trait is encouraged to research Motorized Transportation, which allows Tanks. Advanced Flight is also given a higher weight, so that this technology is researched by some AI, but not all of them waste their time on an optional technology that provides no significant benefit.

As with the first slice of the Medieval age, the last slice of the Industrial age includes technologies that are prerequisites of each other, for comparison reasons.

alexman · January 12, 2004, 17:31

This age was balanced assuming that Spaceship and Diplomatic victory conditions are enabled.

In C3C Rocketry was given such an increase in value from the addition of the Mobile SAM and the TOW Infantry that no attempt was made to prevent the AI from choosing this technology as their initial research project of the age.

Scientific civilizations are encouraged to research Computers that allow cheap Research labs.

It was assumed that Militaristic civilizations aim for a military victory and not so much a spaceship victory. This may not be the best approach in all cases, but it gives the AI some extra variety in research, which makes it more difficult for the human to research some technologies that are ignored by the AI. Therefore, militaristic civs are encouraged to research Synthetic Fibers that allow Modern Armor. Stealth is also a priority, which is a low priority item for other civilizations.

Seafaring civilizations are encouraged to research Miniaturization, which allows offshore platforms. They also have Robotics at a high priority, which allows AEGIS cruisers.

The Agricultural trait is given priority to Ecology, to take advantage of cheaper Solar plants, and to put some weight to the upper branch of the tech tree. They also sometimes research Recycling (cheaper Recycling centers) which is a very low priority item for other civilizations.

As with the first slice of the medieval age and the last slice of the industrial age, the last slice of the Modern Age includes technologies that are prerequisites of each other, for comparison reasons.

Rellin · January 12, 2004, 18:55

Has anyone played a game or two with these flavors yet to see how they work out?

alexman · January 12, 2004, 19:09

What do you mean by 'work out'?

I think the above tables show very well how the AI research choices will be influenced. And if you look back a few posts, you will see how the AI trading will be influenced.

Rellin · January 12, 2004, 19:45

Ok, I went back and looked over the tables and stuff and read your posts more thoroughly and yeah, it looks like it will all work out in theory.

Anyway, just wanted to see if it works well in practice from peoples opinions, and yours would surely do Alex, have you played a game and seen that the AI perform well with these research options, and that trade value and all that work out? It's not just that I'd like to see them research with more variety, but also that that variety aids in thier development as a Civ and in turn as a threat to my domination.

Ok, I want to try this out regardless, but I am not sure exactly how to implement this in the editor. Under Civ Advance/Building Flavor I make the catagories you have made up, and then associate the techs with that flavour, and add the additional attributes to the techs you have done. Then for Civ flavors make the same names.

This is where I get kinda confused, what are the % relationships you used between the traits? Like Seafaring tech flavor has 100% to Seafaring Civ flavour, and then 99% to the rest of the Civ flavours?

And how do you go about assigning Civs thier individual flavours, if a Civ is both Religious and Militaristic does it get marked for both those Civ flavours?

Is Generic both a tech and civ flavour or just a tech? If it is also a Civ flavour, what are the rules regarding applying to a specific Civ?

I see you added Naval Transport to Alpahbet, and I am not seeing how to do that in the editor.

And finally I don't understand exactly what you did with the banana resource to assign additional weight to the tech, not why, but how you did it in the editor.

Really sorry if I am asking too many questions, or sound dumb, but I want to try this out, and am a bit confused still.

Also, I have another question, if I were to remove ToE wonder completely and make Philosphy not give a bonus tech, how much would that affect the work you have laid out here?

alexman · January 12, 2004, 20:58

Quote:

Originally posted by Rellin
This is where I get kinda confused, what are the % relationships you used between the traits? Like Seafaring tech flavor has 100% to Seafaring Civ flavour, and then 99% to the rest of the Civ flavours?

Yes, 99% to everything else.

Quote:

And how do you go about assigning Civs thier individual flavours, if a Civ is both Religious and Militaristic does it get marked for both those Civ flavours?

Yes, it would get both. It would work, because the relevant flavor relationship would be the maximum of the two for each tech.

Quote:

Is Generic both a tech and civ flavour or just a tech? If it is also a Civ flavour, what are the rules regarding applying to a specific Civ?

All civs get the Generic flavor, unless they have two other flavors. If they have just one other flavor, they get the Generic flavor and the other. That's how I did my tests.

Quote:

I see you added Naval Transport to Alpahbet, and I am not seeing how to do that in the editor.

You have to modify the Curragh unit to have the AI Naval transport flag in the editor. Check the 'unload' flag first. This change has already been voted in the mod elsewhere.

Quote:

And finally I don't understand exactly what you did with the banana resource to assign additional weight to the tech, not why, but how you did it in the editor.

Create a new resource. Give it the GOOD_Bananas civilopedia entry. Check it as a strategic resource with an appearance ratio of zero, and select the required tech. Repeat for each technology that requires a dummy resource.

Quote:

if I were to remove ToE wonder completely and make Philosphy not give a bonus tech, how much would that affect the work you have laid out here?

Removing the ToE would affect AI research choices, yes. To get the same AI research priority as with the Wonder, simply flag the 'Allows Bridges' and 'Allows Sea Trade' abilities in the editor for Scientific Method. Removing the free tech from Philosophy would not affect AI research choices.

I have not tested these modifications in a real game, but my tests involved playing an actual game with 30 AI civs for about 30 turns, so I expect things to work smoothly. I will post a mod with the above changes so people can test it out before the real AU game.