Civ4 AI Survivor: Season Four

[Borsche]

The peacenik AI's avoid Rifling mainly because of its flavors - they are 10 Military, 1 Production, 1 Science. For comparison, Gunpowder is 9 Military, 4 Production, 4 Science, Assembly Line is 3 Military, 10 Production, 3 Growith, Industrialism is 2 Military, 10 Production, 3 Gold, 4 Growth. The few non-Military flavors Rifling has is not enough to make it a priority even for those that do have Production or Science flavors.

And Elizabeth actually made the Championship match in AI Survivor 1, finishing 4th place, so she's the English leader with the best performance so far.

I don't feel that Bismark deserves such a low place. His greatest weakness is perhaps his peaceweight.

Keeping with this, I messed around with the tech tree to put the respective flavor preferences on each tech: https://i.imgur.com/qJgD06H.png

With the late renaissance/early industrial era you can really see how leaders with a Science flavor really get pulled far away from Rifling:

[yuris125]

How difficult would it be to modify Rifling's weights to make non-militaristic AIs less eager to avoid it?

[Borsche]

How difficult would it be to modify Rifling's weights to make non-militaristic AIs less eager to avoid it?

Really easy. Here's how the tech is defined in the CIV4TechInfos.xml document:

<TechInfo>
            <Type>TECH_RIFLING</Type>
            <Description>TXT_KEY_TECH_RIFLING</Description>
            <Civilopedia>TXT_KEY_TECH_RIFLING_PEDIA</Civilopedia>
            <Help/>
            <Strategy>TXT_KEY_TECH_RIFLING_STRATEGY</Strategy>
            <Advisor>ADVISOR_MILITARY</Advisor>
            <iAIWeight>0</iAIWeight>
            <iAITradeModifier>10</iAITradeModifier>
            <iCost>2400</iCost>
            <iAdvancedStartCost>100</iAdvancedStartCost>
            <iAdvancedStartCostIncrease>0</iAdvancedStartCostIncrease>
            <Era>ERA_RENAISSANCE</Era>
            <FirstFreeUnitClass>NONE</FirstFreeUnitClass>
            <iFeatureProductionModifier>0</iFeatureProductionModifier>
            <iWorkerSpeedModifier>0</iWorkerSpeedModifier>
            <iTradeRoutes>0</iTradeRoutes>
            <iHealth>0</iHealth>
            <iHappiness>0</iHappiness>
            <iFirstFreeTechs>0</iFirstFreeTechs>
            <iAsset>32</iAsset>
            <iPower>12</iPower>
            <bRepeat>0</bRepeat>
            <bTrade>1</bTrade>
            <bDisable>0</bDisable>
            <bGoodyTech>0</bGoodyTech>
            <bExtraWaterSeeFrom>0</bExtraWaterSeeFrom>
            <bMapCentering>0</bMapCentering>
            <bMapVisible>0</bMapVisible>
            <bMapTrading>0</bMapTrading>
            <bTechTrading>0</bTechTrading>
            <bGoldTrading>0</bGoldTrading>
            <bOpenBordersTrading>0</bOpenBordersTrading>
            <bDefensivePactTrading>0</bDefensivePactTrading>
            <bPermanentAllianceTrading>0</bPermanentAllianceTrading>
            <bVassalTrading>0</bVassalTrading>
            <bBridgeBuilding>0</bBridgeBuilding>
            <bIrrigation>0</bIrrigation>
            <bIgnoreIrrigation>0</bIgnoreIrrigation>
            <bWaterWork>0</bWaterWork>
            <iGridX>12</iGridX>
            <iGridY>4</iGridY>
            <DomainExtraMoves/>
            <CommerceFlexible/>
            <TerrainTrades/>
            <bRiverTrade>0</bRiverTrade>
            <Flavors>
                <Flavor>
                    <FlavorType>FLAVOR_MILITARY</FlavorType>
                    <iFlavor>10</iFlavor>
                </Flavor>
                <Flavor>
                    <FlavorType>FLAVOR_PRODUCTION</FlavorType>
                    <iFlavor>1</iFlavor>
                </Flavor>
                <Flavor>
                    <FlavorType>FLAVOR_SCIENCE</FlavorType>
                    <iFlavor>1</iFlavor>
                </Flavor>
            </Flavors>
            <OrPreReqs>
                <PrereqTech>TECH_REPLACEABLE_PARTS</PrereqTech>
            </OrPreReqs>
            <AndPreReqs>
                <PrereqTech>TECH_GUNPOWDER</PrereqTech>
            </AndPreReqs>
            <Quote>TXT_KEY_TECH_RIFLING_QUOTE</Quote>
            <Sound>AS2D_TECH_RIFLING</Sound>
            <SoundMP>AS2D_TECH_MP_RIFLING</SoundMP>
            <Button>,Art/Interface/Buttons/TechTree/Rifling.dds,Art/Interface/Buttons/TechTree_Atlas.dds,3,7</Button>
        </TechInfo>

Just boost the flavor weight for science, or add more flavor weight to it like production, gold, culture, etc.

[Fluffball]

Gotta nominate Alpha Centauri to do sometime on your Civ Fridays, post somewhere here when you do and I'll follow along.

I'll second that, and also toss in the suggestion to give GalCivIII another try. I fired it up a couple days ago and got completely stomped by the beginner AI because I hand't realized how many improvements they've made. Even a couple years after launch I retried it and they had massively improved the AI, but as of right now it seems like it actually plays the game competently. 

The AI was invading me, which is nothing special because that's easy to program and I had totally ignored military tech, but one of the AIs was dramatically out-researching me on the lowest difficulty setting. 

[Brian Shanahan]

I am surprised the RNG entry did not do a bit better, considering some of the big upsets this season. Hmmm, I wonder if the RNG did better (on a relative basis) for games where the overall average score was lower? I need to look at the data now that the final results are all in.

I think to get proper data we'd probably have to get at least ten people to run random picks if not more. And then the average of the randoms would give us an idea of how accurate it is.

[Brian Shanahan]

As the Twitch chat pointed out during this game, the best trait pairing in Civ4 is "Financial / Alive"

Can we give a forum shoutout to whoever created the best trait pair in the game?

I think it was Anti-Social Monkey.

[Gustaran]

Just want to point out that, in spite of all the surprises this season, the very worst total picking contest score among those who participated for every single game is ... random.org!  It isn't even close.  Now, part of that might be that players with low scores are more likely to lose interest in the contest, but it looks like randomly throwing darts at a board really isn't a good way to pick a winner, even when things seem unusually insane!

I find that not very surprising.

1. First of all, you compare the score only with people who participated in all rounds. It's probably save to assume that most of these people invested at least a certain amount of time (i.e. read Sullla's s preview of the next round) before making their picks.

2. Seeded leaders made picking this time easier for humans. Yes, there were upsets, but picking according to seeds at least guaranteed a certain number of points.

3. I believe the main issue is the range and the number weight you feed into random.org. Do you allow all possible values or do you limit the "randomness" to a sensible range and if so how much?

Example: Pick the final turn. Obviously random.org will do much worse, if you allow for a number between 100 and 400. Pick a "random" number between 250 and 350 and you are probably getting much better results. Same thing with with victory types: Seems that certain types of victory are much more likely to occur than others (for example conquest vs. diplo). Of course, if random.org doesn't take that into account, it's going to do much worse than any human.

I would find it much more interesting to take the data from the last 4 seasons, figure out the average values and use a limited range to pick a random number close to that average. In addition, it would probably be easy to convert the seed number of leaders into pick probability, so higher seeded leaders have a higher pick probability. Same thing with picking victory type.  Only then use random.org and see how it does against humans.

(that's just my general idea, when it comes to number crunching this forum has probably many members who are vastly more skilled than I am).

[Kofiman]

Hmm.. Checking back, the season 3 game 2 writup called it a "fun and memorable game." Conclusions called it a total snoozefest. Heh.

[Brian Shanahan]

Just want to point out that, in spite of all the surprises this season, the very worst total picking contest score among those who participated for every single game is ... random.org!  It isn't even close.  Now, part of that might be that players with low scores are more likely to lose interest in the contest, but it looks like randomly throwing darts at a board really isn't a good way to pick a winner, even when things seem unusually insane!

I find that not very surprising.

1. First of all, you compare the score only with people who participated in all rounds. It's probably save to assume that most of these people invested at least a certain amount of time (i.e. read Sullla's s preview of the next round) before making their picks.

2. Seeded leaders made picking this time easier for humans. Yes, there were upsets, but picking according to seeds at least guaranteed a certain number of points.

3. I believe the main issue is the range and the number weight you feed into random.org. Do you allow all possible values or do you limit the "randomness" to a sensible range and if so how much?

Example: Pick the final turn. Obviously random.org will do much worse, if you allow for a number between 100 and 400. Pick a "random" number between 250 and 350 and you are probably getting much better results. Same thing with with victory types: Seems that certain types of victory are much more likely to occur than others (for example conquest vs. diplo). Of course, if random.org doesn't take that into account, it's going to do much worse than any human.

I would find it much more interesting to take the data from the last 4 seasons, figure out the average values and use a limited range to pick a random number close to that average. In addition, it would probably be easy to convert the seed number of leaders into pick probability, so higher seeded leaders have a higher pick probability. Same thing with picking victory type.  Only then use random.org and see how it does against humans.

(that's just my general idea, when it comes to number crunching this forum has probably many members who are vastly more skilled than I am).

On point three for the leaders and victory type I pick all possible outcomes (including nobody dies), I only exclude the winner pick from second, and both from first to die.

On wars I pick from 1-25, one is too low (probably a better bound is 7-20), and on winning turn 200-500 (which I think is pretty reasonable), except when the RNG picks a time victory, that is an automatic turn 500.

[GreyWolf]

I've replayed the final game. I forgot to activate the cheat code, therefore I simply used the initial espionage (I didn't add more in the course of the game) and an occasional spy in order to see unit movements. Here's the brief report:

You can still see similar settling patterns in the north, with some deviation in the south



Stalin had some early barbarian troubles, causing his capital wheat to be pillaged at one point. He also researched Mysticism late, which had a negative effect on his score. He did, however, conquer a barbarian city between him and Mongolia.



As Charlemagne built the Pyramids, Willem founded Judaism and then declared on Protective score leader Gilgamesh. This was clearly a wrong decision.



Willem being in a war prevented Judaism from spreading to Sumeria, and he didn't manage to spread it to anyone else properly, either. This huge siege-less Dutch stack also failed to capture Lagash, despite a lot of blood being spilled on both sides.


 
A lot of Sumerians and Dutch died pointlessly. However, the entrance of Charlemagne in the war signified a breaking point.



Kublai also joined the war, but his effort to take Delft fell short.



Slowly, Dutch defenses started to break down. Darius was busy bulding wonders, sneering at all this conflict and bloodshed.



Stalin distracted Kublai from further gains by starting a war on Mongolia.



As the Dutch core cities collapsed, Willem launched an excursion into Mongolia. He captured Old Sarai, then returned it to Kublai for a peace treaty, but it was still too late for Netherlands.



Ghuzz, an initially Barbarian city, was captured by Stalin from Mongolians. He, however, had a hard time holding it against Sumerian culture.



Sumeria and the HRE clearly would divide Netherlands between themselves.



Stalin's efforts amounted to nothing, as Tiflis was recaptured by Mongolians. who also added the border town of Yakutsk to their domains.



Charlemagne was the leader to eliminate the Dutch from the game. Although the AI doesn't appear consider diplomacy in its religious choices, this was definitely a return to previous seasons for Willem. His pointless aggression on Gilgamesh did absolutely nothing for his civ but leaving it vulnerable to other attacks. Meanwhile, take a glance on what Darius was building and researching. If other civs' tech fit the year 1060AD, Darius was clearly running away with it.



The division of Netherlands:



Another pointless Russian-Mongolian war ended with no cities changing hands. These wars, perhaps partially explainable by different religions of the two civs, dragged them down.



Darius proceeded to fire a GP golden age, then build the Taj Mahal on the next turn, as the two territorially greatest nations came to an epic conflict. This suited the technologically advanced, but militarily weak Persians just fine.



Despite Gilgamesh's higher score, Charlemagne appeared to have the larger army, probably because he lost less men in fighting the Dutch.



At the end of the Sumerian-HRE war, the HRE solidified its control over most former Dutch lands.



Kublai, finally, attempted to stop the runaway by declaring war on Darius. However, Darius at the time researched Gunpowder, meaning that, with Military Tradition, he had Cuirassiers to Mongolia's Knights, holding a slight military advantage. Furthemore, Kublai didn't appear to bring enough siege, stalling at Gordium (I entered the WB in order to get a proper look at the whole map).



Stalin then decided it's a good idea to attack Mongolia, again. The resulting Russo-Persian attack eventually annihilated Mongolia's army at Gordium.







As you can see, Darius had the sense to research Grenadiers. Meanwhile, Charlemange attacked Sumeria, again.



Desperate, Kublai gave away Samarqand in peace to Darius, who, content, resumed his rabid teching.



Stalin, rather unwisely, signed peace with Mongolia, as the HRE slowly appeared to prevail over Sumeria, despite the latter having the same cuirs-vs-knights advantage.



New Russo-Mongolian border.



Stalin declared war on Charlemagne, sent, as far as I saw by spies and WB, a single Grenadier to HRE territory, then signed peace. What was Uncle Joe thinking? (He probably was bribed by Gilgamesh, but still).



Darius reached Modern Era graphics in 1550. Don't worry, he did have Rifling at this time...



Chuck's rather... diverse army composition, which, however, slowly ate the more advanced Gilgamesh (who even did the correct thing and did have rifles at this point). The HRE used the "sheer numbers" strategy, and it worked.



This says it all:



When HRE researched rifles itself, it became military equal with Sumeria, and the writing was on the wall:



No dplo win this game:



Don't be alarmed by Persepolis reaching Legendary, Darius did not pivot to ineffective cultural attempts this time. He was still ferociously researching:



At this point, Kublai decided to imitate playoffs Roosevelt and declare on technologically advanced Darius with nothing better than rifles and cavs:



The result was predictable:



Welp, that's it:



Darius didn't enage in any "incomplete spaceship" shenangians and won Space at turn 341, 1921 AD.

This time, out of two economic leaders, one self-immolated by unpopular faith and pointless aggression, but the other played the typical economic success game.