AI citizen focus change #11908

EmperorPinguin · 2024-07-03T08:08:49Z

Reworks food prioritization, to become more 'gradual' instead of exactly +2 food surplus.

Reduces value of gold to 1/2 of production, as is applicable during most of the game

Increases value of science, as the great scientist pionts are implicit in their evaluation.

Increases value of Faith. Currently detrimental to AI (they spend faith on inquisitors preventing space victory), but more helpful to humans.

EmperorPinguin · 2024-07-18T19:23:08Z

@tuvus, if we're in negative happiness and we've got a 1-pop expand city with e.g. Old Faithful, I think in many cases it's best to work the happiness so our other cities can continue growing. Similary, if we're losing a lot of gold per turn, we need all our cities' help to avoid units disbandment. I often work gold tiles in small pop expand cities, especially if I'm both in negative happiness and negative gold. The AI rightfully values gold lower in small cities (you don't want citizens to get distracted by camp-luxury tiles early on), but shouldn't be completely prevented from working gold/happiness if need be.

tuvus · 2024-07-19T00:59:35Z

core/src/com/unciv/logic/automation/Automation.kt

        if (city.civ.getHappiness() < 0) {
-            // 75% of excess food is wasted when in negative happiness
-            yieldStats.food /= 4
+            yieldStats.happiness *= 2
+        }


Can you put this in the duplicate if-statement above?

tuvus · 2024-07-19T01:02:56Z

core/src/com/unciv/logic/automation/Automation.kt

+            if (city.civ.getHappiness() < 0) // slotting Democracy specialists is an easy fix to happiness problems
+                yieldStats.happiness *= 2


If you want to do something like this, look at line 57 for the food-related unique. Otherwise you are just guessing that we have the democracy policy unlocked.

Actually, we already have this implemented, line 61-63 accomplish this.

tuvus · 2024-07-19T01:05:17Z

core/src/com/unciv/logic/automation/Automation.kt

@@ -61,6 +61,9 @@ object Automation {
            for (unique in localUniqueCache.forCityGetMatchingUniques(city, UniqueType.UnhappinessFromPopulationTypePercentageChange))
                if (unique.params[1] == "Specialists" && city.matchesFilter(unique.params[2]))
                    yieldStats.happiness -= (unique.params[0].toFloat() / 100f)  // relative val is negative, make positive
+            yieldStats.science *= 1.5f // we want to be working scientists


This is more of a player preference, can we leave this out?

The base value of science is pretty low. We could increase it to make the AI work scientist slots, but this would lead to them getting stuck working trade posts and Krakatoa. The 'cleanest' way to properly implement this, is to make the AI look up the abilities of the great person corresponding to the great person points and value the slots accordingly, but is a bit too complex for me to implement. It seems multiplying the value of science by 1.5 is approximately good enough to let the AI rank scientists as the highest of the specialists.

Good play shouldn't require a player preference. If Civilization VI is implemented, a more sophisticated algorithm needs to be in place, as there the specialist slots don't produce great person points. Also, in Civ VI, the value of a great person point is highly dependent on the specific great person you want to get, rather than the category of great person.

Great scientists in the base game are generally regarded as the best great persons. In Rekmod, each type of great person is in their own gpp pool, while manufactories and customs houses are much better, You could argue engineer or merchant slots are the better slots, but when I visit the Lekmod discord, people there are obsessed with discussing scientist management, so I think it holds true across base rulesets.

This is more of a player preference, can we leave this out?

@tuvus, what's your proposal for letting the AI work the right specialist slots?

The prerequisite questions are:

How do we figure out the best research slot?

Do we want all the AI Civs to choose one specialist type?

Do we want to choose a specific strategy that we deem best and try to get the AI and/or players to use it?

For the first question, if we want to dynamically determine this we'd need some sort of unique evaluation function that can rank the uniques together given the current game state and give us a value. It seems like we are having quite some trouble with this. Choose the best one for G&K or some other version would not play well with mods which could re-balance or completely change them.

For the second and third questions, each nation has their own personality (implemented in #10939), meaning we encourage them to be different from each-other. So why would we try and override that? Making decisions for players doesn't tend to go well, this PR for example was the result of a lighter version of that.

For the first question, if we want to dynamically determine this we'd need some sort of unique evaluation function that can rank the uniques together given the current game state and give us a value. It seems like we are having quite some trouble with this. Choose the best one for G&K or some other version would not play well with mods which could re-balance or completely change them.

True, and uniques also tend to get renamed quite a bit. The simplest (but not fool-proof) way to do this would be to adjust the evaluation of science, as the value of GS points is implicit in it.

For the second and third questions, each nation has their own personality (implemented in #10939), meaning we encourage them to be different from each-other. So why would we try and override that?

If I play test with the current implementation, a custom Default Civ does show to rank scientists as the highest of the specialists, but doesn't work all of them all the time. Songhai keeps prioritizing merchants, and India keeps prioritizing artists (I haven't tested for every civ).

Making decisions for players doesn't tend to go well, this PR for example was the result of a lighter version of that.

The default focus is currently not very good, so I've got to manually assign all my citizens. From playtesting, it appears this PR solves some of it, but I'll likely give it more fine-tuning in the future.

tuvus · 2024-07-19T01:07:40Z

core/src/com/unciv/logic/automation/Automation.kt

+        if (city.isPuppet) {
+            // Puppets need to produce gold, but no great merchants
+            yieldStats.gold *= 2
        }


Can we also leave this out? Setting a the city focus is a bit different than hard coding some values behind the scenes.

tuvus · 2024-07-19T01:08:55Z

core/src/com/unciv/logic/automation/Automation.kt

@@ -39,7 +39,7 @@ object Automation {
            gpp = specialistInfo.greatPersonPoints.sumValues().toFloat()
        }
        gpp = gpp * (100 + city.currentGPPBonus) / 100
-        rank += gpp * 3 // GPP weight
+        rank += gpp * 1 // GPP weight


What exactly is the reasoning for this? The Automation doesn't use specialists very often in the first place.

The AI currently doesn't distinguish between types of great person points, and doesn't know well how to use the ones they produce, which makes a calculation like this not very useful. I think it's best to set it to a low value, to act as a 'tiebreaker' between an engineer slot and an unimproved hill.

tuvus · 2024-07-19T01:11:20Z

core/src/com/unciv/models/stats/Stats.kt

@@ -148,9 +148,9 @@ open class Stats(
    fun applyRankingWeights() {
        food *= 14
        production *= 12
-        gold *= 8 // 3 gold worth about 2 production
+        gold *= 6 // 2 gold worth about 1 production


The AI generally has a problem with obtaining gold. This is partly because there aren't many gold opportunities early on. But reducing the value of gold wouldn't help.

It's not just the AI who uses AI citizen assignment, but also many humans do it. In the early game, the most efficient use of gold is via unit upgrades, which is a gold:production conversion of 2:1. Late in the game, with stock exchanges built, the conversion of base gold to base production is more like 2*(1+0.1+0.1+0.15+0.2) : 1*(1+0.25+0.25+0.25) = 1.77:1, or even 1.19:1 with Professional Army/Pentagon. It makes sense to scale the value of gold with era, but if I'm not mistaken the AI already takes modifiers from buildings into account.

The reason AI loses as much gold as it does, got more to do with how it picks city constructions, and by building roads without connecting cities. Improving their worker management might help quite a bit, but I'm not sure how to do so.

EmperorPinguin · 2024-07-24T09:56:28Z

@tuvus I took your comments into consideration, is it good like this? (this won't be the last PR in the series, but I need more real-game playtesting for new additions)

tuvus · 2024-07-25T02:33:37Z

If we leave out the science * 1.5 then I will probably approve.

EmperorPinguin · 2024-07-25T09:02:40Z

If we leave out the science * 1.5 then I will probably approve.

I removed it, is it better this way?

tuvus

Having gold be valued the lowest, even below faith probably isn't the best, but we can change that later if that is the case.

Otherwise I think this can go through now.

EmperorPinguin · 2024-07-26T07:05:57Z

Having gold be valued the lowest, even below faith probably isn't the best, but we can change that later if that is the case.

I prefer a point of faith over a point of excess gold, and a point of culture over a point of faith. As I said before, the conversion of gold to production is 2:1 via unit upgrades, while the conversion of faith to production is 2:1 via holy warriors. Holy warriors isn't the best faith faith sink, via e.g. religious buildings the conversion is closer to 1:1 (depending on how you do the math, but with an upfront investment into spreading the religion). I tend to prefer a point of culture to a point of faith, but the exact conversion depends on your policies, number of cities, and wether you're growing into good tiles.

Science is harder to rate, as it's a non-substitutable yield, and over half the value of a scientist slot is in the scientist point generation. There are three ways in which to assign citizens, which are 1) regular tiles (farms, mines, pastures, unimproved tiles), 2) specialists (most notably scientists, but other types can be strong as well), 3) trade posts. If we make the value of scientist points implicit in the value of scientist slots (as is currently the case), we need to set it rather high to make the AI prioritise scientists over engineer slots, and over the lower-yield ones among the regular tiles. Doing so, however, will result in an over-evaluation of trade posts, which is bad. According to my sheets calculations, a value of 9 should just be enough to to strike the balance, but it isn't the ideal approach imo.

Note that faith can be converted into science at a rate of approximately 1:10 by purchasing great scientists in late game, and culture is converted into science via the Rationalism tree. The value of gold is amplified by science advantage, as upgrade timings allow you to more effectively convert the advantage into a win. We could set the value of gold, faith and culture according to that of science, but for practical purposes of tile assignment, it's better to set it accoridng to the value of production.

EmperorPinguin · 2024-07-26T07:13:18Z

The best way to gain gold is to grow pop, connect trade routes and to build markets, banks, etc. not to keep stuck working merchants slots and plantations, which hurts the development of our empire. I think it's best to value yields differently for 1) citizens specialist assignement, 2) citizen non-specialist assignement, and 3) building construction assignement. Each one will also have to be weighed according to stage of the game for optimal performance, which is maybe a bit complicated due to mod-compatibily and which victory types are enabled.

EmperorPinguin added 30 commits June 23, 2024 17:26

AI behaviour changes

7ed4c0d

Merge branch 'master' of https://github.com/EmperorPinguin/Unciv

47184b0

Merge branch 'master' of https://github.com/EmperorPinguin/Unciv

e856a76

Update Automation.kt

b571e76

Update Automation.kt

cee0287

Update Automation.kt

16ae6d9

Update Automation.kt

a769a52

Update Automation.kt

9c8f2e2

Update ConstructionAutomation.kt

777b828

Update Automation.kt

2a4f9b0

Reverting some changes

33b9fc7

Changes

e311e17

revert changes

cade007

revert changes

318c396

revert changes

012afb5

revert changes

f4d02c8

Update CityLocationTileRanker.kt

cae12b6

Citizen assignment for stat conversion

9040bab

Merge branch 'yairm210:master' into master

a33d7f4

Merge branch 'yairm210:master' into master

e53f887

Update CityLocationTileRanker.kt

3f540a8

Merge branch 'yairm210:master' into master

0e5db36

Merge branch 'yairm210:master' into master

43cf8a3

Merge branch 'yairm210:master' into master

85743fb

Reduce AI settling

a1f5905

Merge branch 'yairm210:master' into master

d061b67

Avoid AI building units when in negative Supply

4257a0f

Merge branch 'yairm210:master' into master

44c8a90

Update CityLocationTileRanker.kt

5f42454

Merge branch 'master' of https://github.com/EmperorPinguin/Unciv

6b49cdc

EmperorPinguin added 6 commits July 18, 2024 09:04

Update Automation.kt

395ac8f

Update Automation.kt

96acd16

Update Automation.kt

9e73243

Merge branch 'yairm210:master' into master

5de758d

Update Automation.kt

6fc1b0c

Merge branch 'master' of https://github.com/EmperorPinguin/Unciv

53057d9

tuvus reviewed Jul 19, 2024

View reviewed changes

EmperorPinguin added 6 commits July 19, 2024 09:57

Update Automation.kt

d5e270b

Update Automation.kt

c34b62b

Update Automation.kt

611378b

Update Stats.kt

1e4d88b

Merge branch 'yairm210:master' into master

5863788

Update CityTurnManager.kt

c896436

EmperorPinguin marked this pull request as ready for review July 24, 2024 09:48

EmperorPinguin added 3 commits July 25, 2024 10:51

Remove specialist science modifier

8bb6b37

Merge branch 'yairm210:master' into master

0d3d37b

Merge branch 'master' of https://github.com/EmperorPinguin/Unciv

a73bf52

tuvus approved these changes Jul 25, 2024

View reviewed changes

yairm210 merged commit 099a238 into yairm210:master Jul 27, 2024
4 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

AI citizen focus change #11908

AI citizen focus change #11908

EmperorPinguin commented Jul 3, 2024 •

edited

Loading

EmperorPinguin commented Jul 18, 2024

tuvus Jul 19, 2024

tuvus Jul 19, 2024

tuvus Jul 19, 2024

tuvus Jul 19, 2024

EmperorPinguin Jul 19, 2024

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 22, 2024 •

edited

Loading

tuvus Jul 22, 2024

EmperorPinguin Jul 23, 2024 •

edited

Loading

tuvus Jul 19, 2024

tuvus Jul 19, 2024

EmperorPinguin Jul 19, 2024

tuvus Jul 19, 2024

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin commented Jul 24, 2024 •

edited

Loading

tuvus commented Jul 25, 2024

EmperorPinguin commented Jul 25, 2024

tuvus left a comment

EmperorPinguin commented Jul 26, 2024 •

edited

Loading

EmperorPinguin commented Jul 26, 2024 •

edited

Loading

		if (city.civ.getHappiness() < 0) // slotting Democracy specialists is an easy fix to happiness problems
		yieldStats.happiness *= 2

AI citizen focus change #11908

AI citizen focus change #11908

Conversation

EmperorPinguin commented Jul 3, 2024 • edited Loading

EmperorPinguin commented Jul 18, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

EmperorPinguin Jul 19, 2024 • edited Loading

Choose a reason for hiding this comment

EmperorPinguin Jul 19, 2024 • edited Loading

Choose a reason for hiding this comment

EmperorPinguin Jul 22, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

EmperorPinguin Jul 23, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

EmperorPinguin Jul 19, 2024 • edited Loading

Choose a reason for hiding this comment

EmperorPinguin Jul 19, 2024 • edited Loading

Choose a reason for hiding this comment

EmperorPinguin commented Jul 24, 2024 • edited Loading

tuvus commented Jul 25, 2024

EmperorPinguin commented Jul 25, 2024

tuvus left a comment

Choose a reason for hiding this comment

EmperorPinguin commented Jul 26, 2024 • edited Loading

EmperorPinguin commented Jul 26, 2024 • edited Loading

EmperorPinguin commented Jul 3, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 22, 2024 •

edited

Loading

EmperorPinguin Jul 23, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin Jul 19, 2024 •

edited

Loading

EmperorPinguin commented Jul 24, 2024 •

edited

Loading

EmperorPinguin commented Jul 26, 2024 •

edited

Loading

EmperorPinguin commented Jul 26, 2024 •

edited

Loading