GOTM9 - The Japan Campaign site is now active!

[cracker]

The mathematical basis for the comparative length of the streakiness would be the divided size of the probabilty space.

The outcome function is using the terrain adjusted combat statistics ratio from the combatants to determine where the dividing point should fall in the probability space. In the case of the war chariots from the first edo table, the dividing point was at 61.5%. With a perfect RNG, 61.5% of the outcomes would favor the attacking war chariots. The chances that the second outcome would favor the war chariots would be 61.5% vs 38.5 for the Japanese units. The ratio there is 2 to 1. So it is a little less than twice as likely that a streak of wins will continue to favor the war chariots as opposed to switching to a win for the Japanese. (the actual expected average ration would be about 1.65) This is an independent test, but it does indicate that we should expect winning streaks of consecutive events that on the average are twice as long for the war chariots as they are for the Japanese.

If you took this to the extreme and just had the combat trade hits it would go something like 0-1-1-0-1-1-0-1-1-0-1-0-1-1 you could see that the forced circumstances would be the expected 1.6 to 1 ratio.

If the RNG is truly random but will still pass the uniform distribution test, then you would expect a similar prevalence of streaks on the high and low side. The streaks would randomly occur, but on the average their length would be in ratio to the probability space.

There are some test statistics that get applied to RNGs to measure things like streakiness and the nominal sample size that can be expected to produce an average sample set. I am not too familiar with these factors but I was still very amazed at how often the actual stroke results watched with the expected statistics.

Yes that one table is a small sample of only 34 events but there were 17 other table event sets that all matched fairly close to the expected statistics. I only formatted 8 of those event tables for the website because I thought a little of this topic would be interesting but if we went too far towards this statistical analysis stuff, then we would just alienate the majority of the readers and invite diatribes from the fanatics who feel the combat event system is totally screwed up.

I think it works great for standard slugfest land units like these examples it just starts to fall apart when you begin to add in units from the other genre (bombardment, naval, and air). But that is another set of topics that belong in another forum.

 

[Heikki]

Cracker,

While the basic probablistics that you present above are simple and valid, the mathematical bridge I asked you to build between the probability of a single weighted flip and the probable length of a streek is still not there.

But, as you said, such discussion is not of interest to the majority of the forum, so let's leave it there.

However, I find it odd that you are assuming such low quality of the RNG in Civ3. I'm not sure what exactly you are refering to with the uniform distribution test. But if you mean that the RNG should produce a uniformly distributed set of numbers for any given set, well, that just wouldn't be random.

If you draw sets of 100 numbers from a RNG, one after another, and each of them is reasonably uniformly distributed, the RNG is *not* a good one at all. It should produce a strongly weighted set every now and then.

If you are talking about very large sets (in the region on million events), the distribution would indeed be uniformly distributed by a good RNG, but that is completely irrelevant here, since no conclusions at all affecting macro-strategies could be drawn.

Anyways, I suggest that you try this: when making your indeed useful tests and calculations, assume that the RNG is ideal; that is, every event is *completely* independent of previous events. I'm sure that in the long run you will find this to be the case.

Or maybe we'll simply agree to disagree here. Anyways, appreciate your work & strategic tips.

Heikki

 

[Cartouche Bee]

Well I finally had a chance to look at this work of art! Great stuff! Can't argue with success, I know that part way through those Japanese wars I stopped producing war chariots cause there were too many places that they could not go without roads, I switched to horsemen and then knights. I also don't know if I totally agree about not topping up armies, although the naval transport problem is a good argument.

Still haven't had one of those corona bugs in any of my games, I'm jealous.

Oh and OT, I do agree about the streakiness of the RNG, too streaky for a true RNG.

 

[civ_steve]

Very impressive review!

I'm curious about some of your earlier decisions that led to your starting position. I didn't save many times in the BC period in my game, but my 30AD save shows that Japan had already formed cities up to his NorthWest tip, and on the small and medium sized islands between Japan's and Egypt's starting continents. My first problem was getting control of these islands before I could even set my sights on Japan's main continent. I didn't make a beeline to map-making, I was delayed by barbarians, and maybe Japan in my game expanded more aggresively. I'm interested in how some of these early decisions, and random events, may impact the game that follows.

Also, would you say that the small amphibious force was taking advantage of the AI? I'd think that against a human player, that small force would last maybe 3 turns, and the horses restored in 5. (Perhaps the annoyance factor is sufficient reason to employ them, along with denying horse resources for 5 turns.)

 

[cracker]

Civ Steve.

I can't say that the samll task force was taking advantage of the AI, but I was totally shocked at some of the AI behaviors I was able to see because of using the multipoint attack.

I mentioned two of these observations in the Satsuma battle page at:
http://www.civfanatics.com/doc/civ3/cracker/japan_campaign/satsuma.htm

where I had the small task force of 4 units sitting just next to the town of Satsuma. Instead of attacking this task force and hurling it back into the sea. The AI did basically nothing during its first turn after the troops landed. In the second turn, the AI pulled the offensive units (an archer and a horseman) out of the cities that surrounded the landing zone and insteadd on attacking the landing zone, the AI moved the offensive units back toward Izumo to atatck the stack made up primarily of war chariots. I think that some of this has to do with a calculation that the AI may process to assess where it may find the weakest possible target. The task force at the landing zone was guarded by a Spearman and a Swordsman, while the seven remaining war chariots in the nearby Izumo attack where basically exposed.

I think the offensive AI may have viewed the chariots as an easier target and just forgot about its defensive alter ego.

I would have observed the Galleys that were moving along the northern coast and would have positioned offensive units to be able to strike any landing force when they stepped off the boat. When the units landed and where unfortified they would be at their weakest. I think this is part of the difference in the AIs tactics that tend to rely of just standing there and duking it out versus using an active defense.

Knowing that the AI has basically two combat brains, with the offensive and defensive halves, it is easy to see how the offensive half looked for the easiest target and ignored its role in reducing the task force because that little stack may have been a less successful attack. One half the brain controls the offensive units and the other half the brain controls the defensive units.

Another thing the AI does not seem to do well is understand the attack range of units. The AI will move to attack a weakened unit or a worker that is not stacked with a strong defense unit, even if that weak target is covered by the attack ranges of a large number of stronger attack units.

Another AI tactic gap that clearly showed up was the way the AI used its catapults (probably applies to other bombard units as well). When I advance the two attack stacks into squares immediately next to Kyoto, the AI had one movement turn between when the units got next to the town and the time when I could actually attack. If you or I were under attack by a stack of units, as sonn as they came into range of our bombard units we would start pounding them to reduce their strength. Instead, the AI held its fire and did not fire any of the three catapults that it had available. When I attacked the city, the catapults each took their one free-shot, but they should have already taken three other full fire shots at the advancing units.

There were many very interesting events in these attacks when I was able to look back at my notes and see what happened.

 

[Franses]

Cracker,

A great site and a story well told. Like Heikki it made me register here (generally I post on Apolyton).

Like Steve I am very interested in the way you reached the position in 170 AD. I looked back in my old GOTM09 saves and found out that at around 170 AD I was just at the point to built a city on one of the other Islands.

I replayed the game optimizing for expansion (using version 1.29 but I do not expect that to make a difference). I was able to settle two cities on the Island of Japan and a couple on the three other Islands but I did not get as far as you nor did I get the military and temples that you had.

Did you get a Settler from a goody hut early in the game? If not, I have to reevaluate my CIV3 skills (very frustrating ). Could you elaborate a bit on how you reached this great position?

 

[Zachriel]

Great and very detailed job. What I want to know is what the doctoral committee thought of your magnus opus? Certainly you have earned the title of Dr. Cracker!




German Valor
http://www.zachriel.com/gotm8/

 

[cracker]

There was a discussion about the impact of V1.29 when Exanguination asked if it was OK to switch to V1.29 in the middle of his/her GOTM9 game, and he/she thought I was just being a jerk when I posted that switching to V1.29 was not a bright idea in this case.

I tested some of the files and found that V1.29 increased the human research cost and slowed the tech development of both the human and the AIs. For self researched items, V1.29 added 3 to 5 turns and delayed game progression. It also slowed delievry of items out of the Great Library in the mid game because the AI tech research cost shifted from 8 to 10 with the patch.

In the early game, growth production was effected by whether you had the cow in the start city or not plus whether you use the key settler from hut strategy. Whenever I found a hut, I would make sure all production was shifted to either the pyramids or the palace or something else other than a settler. If a settler was on the loose I would wait a turn to pop the hut until the existing settler was planted into a city. Otherwise you turn off the chance that a hut may produce a settler. Of the four starting huts, I did not get barbs out of any of them even though I had several camps that popped up. I don't think that micromanaging the settler popping has much benefit after two settlers because you may be at or above average city numbers at that point.

Heliopolis came from a settler in the north mountains and I had one more settler from a hut as well before I think we pushed into parity in towns. I got bronze working from a hut and other than that, we only researching Writing, Mapmaking, and Literature in that order. I shut down research completely and just went naked until we popped into the 2nd age and then researched engineering while the other civs worked to give us Cathedrals and Pikes through the library.

I have started some notes on a technique called "short rushing" that I used whenever I could. I am still working on the details of using the technique well but basically it looks at rushing 19 shields whenever possible and then switching production to something larger. So if you have a town that has at least two people on emporer level, and if that town has one shield in the bin, you can set it to make a spearman and then rush the spearman full. But you do not complete the spearman, you switch over to galley, settler, or temple instead depending on the population. In other cases you might hack a forest for 10 and then rush the remaining 19. Without short rushing, more shields are wasted because any current production gets overlapped with the rush quantity and basically gets thrown away. I know that short rushing gave me some galleys, temples and a few settlers that would have taken longer. If there had been less jungle nad more forest this would have even gone quicker.

 

[abla]

How close do you build your cities?
From the little bit of the Egyptian homeland visible on some of the maps you look to have a very dense build?
By the time I got mapmaking in my game Japan had already filled up the north and west of his island so there was nowhere left to land without upsetting him.
I ended up competing with him for the jungle island to the north and working on culture flipping to get his cites out of there.

 

[cracker]

Abla,

I don't think that my build was that dense at all.

Here's a link to the original spoiler thread where CB and I compared positions at the same point in time near 770AD:

http://forums.civfanatics.com/showthread.php?postid=368300#post368300

If you look at the mini-maps in those images, you can see that CB had 1.5 to 2 times as many cities as I did. In the Egyptian homeland, CB easily had twice the towns in the same area. I think CB's postion was probably more powerful while I was probably spread out in more of a land grab and using the whip every 20 turns if possible.

I don't know what drives the decision algorithm in the AI programming with respect to settlers, but if you add up the captured units from the position of the Japanese when we begin our war, they had 6 or 7 excess settlers built up in piles in their cities.

I think that getting a warrior or two over on their continent early may screw up their city placement functions even if you cannot actual stop them all together.

The Japanese founded Toyama and Fukushima while our early warriors looked on.

One thing I also found was that the replay for the website did not play exactly the same as the original game. I had not save very many early files until I got to contact with Japan, and then in teh Japan Campaign area I only had about 3 original save files to go back and work from. If you cannot replicate the exact sequence of build items in every city plus teh exact sequence of unit moves for every unit, then the outcomes shift a bit with each passing turn. A couple of the battles went statistically much easier in the original game than they did in the documentary replay to generate the animated images.

 

[Franses]

Thanks Cracker,

This explains things for me. I did not know this "key settler" strategy and therefore probably did not get a Settler early. The first goody hut gave me bronze working, the others barbs.

I also do not use the short rushing you mention. I have seen people reporting this "bug?" and thought it was not allowed in GOTM.

 

[Heikki]

Thanks for the short rushing idea, Cracker. Can't wait to get to test it. Have to finish my current game first though, 'cause the situation is really interesting. I'm letting two of my neighbours wage war on each other by having a ROP with both. Then, when a border city falls, in come my settlers. Julius & Alex are soooo gullible.

Oh, and by the way, I opened a new thread on the RNG in the general discussion forum, in case people discussing it here are interested.

 

[abla]

Looking at that map (http://forums.civfanatics.com/showt...8300#post368300) my city density is actually very similar to yours - the bit that threw me was 3 cities on the little island to the east of homeland - I only had one city there!
Obviously city closeness is not the solution then - oh well, back to the drawing board.

 

[TheSandbag]

Cracker -- This is an OUTSTANDING site! Very detailed, very consice, VERY informative. Keep up the great work!

 

[DavesWorld]

Excellent work Cracker.

Not only did you have complete thought paths worked out for what you wanted and how to go about it, but you also were able to clearly illustrate same with your writeup. A+ from me ... that and a buck fifty gets you a small cup of coffee at Starbucks.

I agreed with the point of your writeup as stated ... to help showcase the kind of thought that must go into Civ to get good results out. You had a clearly defined strategy and set objectives as new information became available to indicate tactical need.

Barracks: you've said you were using Despotism and slavery-rushing to speed production. I still don't understand why you didn't slave-rush a barracks in the military production cities, unless it was for budget reasons perhaps. By taking the extra five or so turns to set barracks up ahead of time (you made contact in 630 BC, and by 230BC you were fully preparing for war, which began in 210AD), you would have increased the effective firepower of all your units, offensive and defensive, by a third (4hp each vs 3hp). This would have also resulted in more Elite (5hp unit) battles, directly contributing to one of your overall strategic goals of generating two leaders from the war. What thoughts did you have for not doing this sometime between 630 BC and the early AD?

Armies: Once you get the first leader, which in almost all circumstances should certainly be used to create an army that will lead to an army victory to activate the Military Academy (enabling you to turn production into an Army), your only limitations on armies are your production rates. Yes you can't upgrade units in an army (which I disagree with, personally, as a function of game rules) ... but especially considering you were in a shooting war already, I would have filled the army and got it involved in the fray immediately. By the time technology renders the army invalid due to lack of upgrades, your Japanese war would have been over and also you likely could have produced one or two armies from the MilAcademy. After producing your second army, that should leave you with three active formations and then enable the Pentagon, which enables four unit armies.

Based on your situation, either three War Chariots or three Swordsmen would have been my choices for the army. The former gives you higher mobility and two attack chances per turn .. the later gives you an absolutely excellent heavy attack unit with three hp-sharing 3/2 units, but one-square mobility and one attack per turn. Considering how bloody the Kyoto battle was both expected to, and finally turned out to, be ... a Swordsman army may well have saved a unit or two in losses. Of course at the expense of the two or three turns of recovery time for the Swordsman army following Kyoto's fall.

Interested to hear your further comments. Good job again.

 

[Beard Rinker]

Cracker,

Great story and an insightfull analysis of an early battle.

In my game, my first leader was used to build a forbidden palace. The Heroic Epic and Military Academy are important but pale in comparison to the forbidden palace IMHO.

By your own telling, you very nearly did not get that second great leader. My guess is had the second one not appeared, not building the forbidden palace would have effected your game outcome much more than not building that first army.

I would be interested in why you find building the first army more important than the forbidden palace. For that matter, I would be interested in anyone's opinion on the matter.

 

[civ_steve]

Thanks a lot for your discussion, Cracker! A few things are clearer now.

I got bronze working from a hut and other than that, we only researching Writing, Mapmaking, and Literature in that order.

I decided to investigate my area first. Knowing that this was archipeligo, I should have beelined to Mapmaking. (On some level, this is extra information that in a truely blind game setting, you would not have.) I instead researched Bronze Working and Iron Working before Writing and Mapmaking, so I was definitely several steps behind.

Whenever I found a hut, I would make sure all production was shifted to either the pyramids or the palace or something else other than a settler. If a settler was on the loose I would wait a turn to pop the hut until the existing settler was planted into a city. Otherwise you turn off the chance that a hut may produce a settler. Of the four starting huts, I did not get barbs out of any of them even though I had several camps that popped up. I don't think that micromanaging the settler popping has much benefit after two settlers because you may be at or above average city numbers at that point.

Awesome! I think I've read about this before, but you've made it quite clear. My barbarian huts gave me Pottery, an empty settlement and two Angry hordes of warriors. Those two sets of warriors, along with the occasional camp that sprang up, tied me up in my early game. Getting a settler or two early on really makes a difference.

Thanks also for the tip on short rushing. All in all, you've really created an informative volume.

 

[Greg Loader]

Dr. cracker,

I'm in awe.

Great Job.

Must go read some more.

 

[cracker]

Davesworld,

The barracks issue is a tough one, the war chariots you see in the game were coming from all sorts of frontier cities, so having a barracks in evry one of them was just not possible. Yes, it would have been nice to have barracks, but I thought I needed other things.

I think the army discussion is a key ongoing debate. Read the info carefully on page:

http://www.civfanatics.com/doc/civ3/cracker/japan_campaign/climax.htm

I think I outline a fairly string case for NOT loading the army full. The trump card in the discussion was really whether I wanted to ever use the army for any purpose ever again. Your discussion about the value of the military acadaemy and production to make more armies is valid but you may need to look at the pace I was pushing in this game.

I did end up making more armies but for the most part left them empty (I posted a pick somewhere just to be cute and show three or four empty armies).

I strongly do not think that armies are a good strategic idea in most cases. For me they slow the game down by depriving my units of movement and attack strokes. Mobility and number of strokes is king. Avoid slugfests whenever possible.

For use against fortified defenders on hill cities that you must take they may be invaluable but I am perfecting the "keep em empty or lightly loaded strategies.

I would never load an army with even one swordsman if I had any horse units available and was not facing Greece or Rome as an opponent (or if the only choice I had was to attack pikes or muskets with swordsmen from way behind in techs). I feel fairly strongly about using NEVER in this sentence.

 

[cracker]

Beard Rinker,

In general, I might agree with your FP thought process but I use the general rule of "Early in the war, build a one unit army; Late in the war build the FP."

The key for me is, do I think there will be enough more combat events to probably generate another leader. If we will be making peace or killing the opponent soon then this could be our last great leader for a while plus the army will just sit useless and probably go obsolete.

I think that a mistake I made in this game was not prebuilding the Heroic Epic to a more complete status before geeting into the war. Having the HE prebuilt is an important part of geeting the most bang for your buck in using the first leader for an army.