this is driving me nuts!

[Morten Blaabjerg]

This is beginning to really get to me...

Of the followingg naval units why the f*** does the AI pick the Galleass as its favourite...? -the galleon is more expensive, but far more durable and effective weapon.... Also, the only ship that the 'instant advice' will recommend to build is the freaking galleass, which is only included for historical color, but was ultimately completely outclassed by modern war galleons. Can it really be right that I have to obsolete this vessel or make it unbuildable just to get around this?

Does the recommendations/build preferences have something to do with the number of units alredy in existence? cost? stats? roles? Think I've tried a zillion things by now, changing cost, att, costs and stats of competing vessels, roles, switching unit slots... still the freakin' galleass is the ultimate thing for the AI. it is clearly the favourite ship for the AI. Can somebody explain me why?

Hulk, Mag, 2, 2.,0, 1a,8d, 4h,1f, 12,2, 4, Phy, 000000000100000
Cog, Phy, 2, 2.,0, 1a,6d, 2h,1f, 8,2, 4, Inv, 000000000100000
Carrack, nil, 2, 3.,0, 2a,16d, 4h,2f, 16,4, 4, Mag, 000000000100000
Caravel, nil, 2, 3.,0, 8a,8d, 3h,1f, 12,2, 4, Nav, 000000000100000
Galleass, Rob, 2, 3.,0, 12a,8d, 3h,2f, 12,0, 0, Min, 000000000100000
Galleon, nil, 2, 3.,0, 16a,20d, 4h,2f, 24,0, 2, Min, 000000010100000

Any tips on how to change the way the AI picks it favourites? Its driving me nuts in this armada scenario. I really want to get the balance right, since a great deal is about ships. The funny thing is I don't recall similar experiences from my hammer scenario, where the AI would build a lot of different type ships, and recommend many different kinds to be build too.

Any ideas or experiences?

 

[Morten Blaabjerg]

Just managed to produce a new result : a carrack !

Amazing !

But I would still really like to know : any (verified, if any) ideas on how the AI picks units to build or recommend building?

 

[Henrik]

If you really dont want the AI to build it, just place it in the carrier slot. The AI wont ever build the unti in that slot (maybe that is the problem with the galleon I don't remember you're units.gif right now...).

 

[Morten Blaabjerg]

Its an idea... -I took care to place the gallon in the battleship slot specifically, so its not that. On closer inspection, the AI do seem to build other ships, but clearly it has its favourites. I just wonder how come it pick such an odd favourite when the galleon clearly is the master of the ocean (I know you favoured the galleasses, Henrik, but I still wonder why)

 

[Boco]

My guess is that the preference for the Galleass stems from its role. Is the AI wanting to be aggressive? What happens if the Galleon is given a role 0? Still, I have no idea why the AI would opt for role 0 over 2.

The example of Carrack preference could be related to role and techs or possibly money. It's the most largest expensive transport. Perhaps in this instance, the AI wanted a transport. Did it build the Carrack in a port near a surplus of land units (of course I'm giving the AI a lot of credit here )? The Hulk and Cog must have been obsolete, but what about the Caravel? Was that a legal option for that AI civ?

 

[techumseh]

I've suspected that what the AI builds is often governed by the number and size of the cities of a given civ. A small civ will build destroyers and subs, but only a large one will build battleships. Or so I hypothesize.

 

[Henrik]

I favoured galleases for two reasons, they where strong enough to bombard effectively and as I recall they where A LOT cheaper than the galleons (what's not visible from your posted section of the rules is that each shield row has more than 10 shields in your scenarios).
Then again I did overvalue the galleases during my play as I falsely assumed they'd be able to take on larger ships as well atleast some of the time. You proved me wrong there as I recall

(for confused readers morten and I playtested this scenario in multiplayer a few weeks ago )

 

[kobayashi]

this may not be directly related but I suspect the AI doesn't take into aco**** firepower/hitpoints when calculating the cost to ability ratio. I always have the same hp and fp for all units if possible.

 

[Morten Blaabjerg]

Thanx, guys! -managed to get a balance together that looks somewhat alright... mainly by adjusting prices and att/def statistics, and a few unit obsolescences, which made sense.

Also fixed another extreme preference, musketeers over arquebusiers, which however is somewhat more reasonable, since the units have different roles in this scenario (but way too extreme, after 12 turns one civ had 76 musketeers (!!) and just 6 arquebusiers (all of these preplaced, none built). Seems the AI preferred to build cheaper longbowmen when wishing an offensive unit, over the better and more destructive, but less reliable and comparatively more expensive arquebusier.

I think you are right koby, about the hp/fp thing. If the AI wants an offensive unit, my bet is it looks at the att value and price of units with that role and not much else.

Think there is something to say too for size, at least as regards production capacity in a city. I.e. if a city has all major improvements and a number of good defensive units, it is more likely to build the most heavy and expensive offensive units available, be it ships or landbased units.

Interesting to see how it plays out in the long run.

 

[Morten Blaabjerg]

Just found this gem of advice in a tip by Mike Daumen at the Scenario League Design Tips. Thought this might be of use to others. It is yet untested, but I will definitely check my units-section for problems like these.

Statistics and Obsolescence

A thornier problem arises when the computer selects units to be built. Many designers are unaware of this bug as it only affects computer-controlled nations. If you examine each nation by playing as a human player, all units will be displayed; only by revealing a computer-controlled city will you notice this problem.

The computer will not build a ground unit under any circumstances when another ground unit with the same role (the number just before the prerequisite advance in rules.txt) and movement allowance (1 or 2+) is available with better attack and defense statistics. The computer ignores cost, special abilities, hit points and firepower. Compare the following examples:

Arquebusiers have attack 2, defense 1, move 1, and role 1 (defense). Regulars have attack 2, defense 2, move 1, and role 1. the computer will never build an Arquebusier if it can build a Regular, since the latter has equal or better attack and better defense. If the Regular's attack was lowered to 1, the computer would then build both units.

Compare the Arqebusier above (2/1) with a Grenadier of attack 3, defense 2, move 1, and role 0 (attack). Because the roles are different, the computer can build both units.

Compare the Arqebusier above (2/1) with a Footsoldier of attack 1, defense 3, move 1, and role 1. Although the Footsoldier has a greater aggregate total of statistics (4 vs. 3), because both of the Footsoldier's statistics do not equal and/or exceed the Arquebusier's, the computer will build both.

Compare the Grenadier above (3/2) with a Lancer of attack 5, defense 2, move 2 and role 0. Because of the difference in movement, the computer will build both even though the Lancer is clearly the better statistical unit.

Compare the Lancer above (5/2) with a Cavalier of attack 4, defense 2, move 3 and role 0. The computer does not distinguish movements over 2, so the computer will not build the Cavalier since the Lancer has better statistics.
When tackling this problem, I typically divide ground units into five types:

units of move 1 with the attack role, including marines, paratroopers, and any assassin-type units that are destroyed after attacking;

units of move 1 with the attack role, a high attack score and low defense score, such as catapults and artillery;

units of move 1 with the defend role;

units of move 2+ with the attack role;

units of move 2+ with the defend role.

Within each type I try to have a spectrum of statistics, and liberally use the pikemen flag (a defense bonus of 50% vs. attack units with a movement allowance over 1) to make further distinctions. Other designers avoid the problem by using the air superiority role (3). I have not done or tested this. In addition, having several units with equal stats but differing hit points, firepower, or special abilities will avoid the dilemma. The key to solving this problem lies in careful examination of cities controlled by the computer, noting which units appear in a city's production window.

 

[Shaka Naldur]

you can make two units exactly the same
the same graphics and one slight difference in the name

that way the ai will build it more often

 

[Morten Blaabjerg]

Anyone who understands how the advisors work?

 

[Morten Blaabjerg]

Sorry for bringing this up again, but I'd really like to figure out how this works. In this case, the AI civs will build both pikemen and halberdiers, which, firstly, sort of disproves Mike Daumens theory...

But apparently the advisors for the player-controlled civ will prefer Halberdiers anyday over Pikemen, despite their obviously better stats in all regards. The advisors will recommend pikemen, but apparently ONLY after suggesting the halberdier, and after at least one or two Halberdiers has been built in the city. I have not tested this thoroughly yet, but it strikes me as extremely curious, and seems to be the case with most cities and civs in this scenario. Even switching the unit slots between the two doesn't help.

As if this wasn't enough, reversing the unit stats doesn't help either!! -it still takes the Halberdier, even if now the advice at least makes more sense... It can't be in the name of the unit can it??

From what I can tell, it seems the advisors pick their odd favourites no matter what. They simply have a mind of their own.
Gonna try switch names and see what happens, but I think this is a little absurd....

 

[Morten Blaabjerg]

Hmmmm.... apparently swithching the names, AND the unit slots one more time did the trick.....

Now it prefers the pikeman as the initial defence unit which makes much more sense... but I still haven't got the slightest idea why?

...it still likes the Halberdier very much, though. After one pike built, it recommends the inferior Halberdier time and time again......

 

[kobayashi]

Can we see the unit stats from the rules.txt?

Also exactly which slots are they occupying?

 

[Henrik]

Is this just the advisor or the AI as well?

If it's just the advisor couldn't you simply include an "Advice.txt" file to fix it?


EDIT: no that's not right, I cannot remember the name though. There is a file you may include to fix the priorities of the advisors rather than the messages that show up (which is what the file mentioned above does).

 

[Henrik]

Got it it's the "CITYPREF.TXT" file (not present in the normal game you have to create it on your own).

 

[Morten Blaabjerg]

This sounds very interesting Tell me more, tell me more !

Initially I got the hunch that the advice given somewhat mirrors the preferences of the AI. From what I tested last night it does look like there something to that. I managed to make the advisor prefer the Arquebusier over the musketeer by switching roles with the musketeer, from attack to defence and vice versa, and it seems the AI civs now build much larger numbers of arquebusiers too.

Before this change, the AI would consistently prefer the musketeer, and use them for its mindless attacks anyway. So what I figured is that even though the AI obviously plays a defensive game, it throws whatever units it has into attacks, it might as well use the better infantry unit to do it. It still builds the more expensive and better defending musketeers, but in much more reasonable numbers.

Still having a lot of trouble making the AI build ships... But this problem is more well known. Gonna take a look at Brown Mans Burden, where I noticed the advisors actually prefer a quite wide range of naval units. Carl Fritz has made the AI work incredibly well in that scenario too, it seems. So the advice and the AI might very well use the same production preferences at the root.

 

[Morten Blaabjerg]

Here's the stats from rules.txt. Between the two rules.txt's there's been numerous changes in the costs of the units, but I tend to agree with Mike Daumen that cost doesn't make any difference, in neither advice nor AI preferences.

The musketeers use the dragoon slot, and the arquebusier the cavalry slot. The pikes and halberdiers use the fighter and bomber slots respectively.

Before the changes :


The advisor consistently prefers the Halberdier over the pikeman, and even still when the halberdiers was made drastically inferior, in all statistics. (see below)

Additionally, the AI will build musketeers in overwhelming numbers, on the behalf of the offensive arquebusier, which is built in much lesser numbers. Adjusting the stats of the arquebusier to a more cost-effective, slightly better ATT-unit, and making the musketeer lesser on DEF (like below) doesn't change preferences noticably, although now the Arquebusier has a more distinct role. Still the musketeer is the clearly preferred unit. Switching roles did the trick (which is clearly not logical, but it does the trick).

After the changes :


The advisors will now prefer the pikeman as the first defensive unit built in a city (think the same holds true for the AI). It still likes the Halberdier a lot, though, even after the changes.

And the change in role for the arquebusiers/musketeers makes the AI build more offensive Arquebusiers over the higher priced defending musketeers (price still doesn't appear to matter, though, I bet I could make the Arquebusier more expensive now with the same results).

 

[Morten Blaabjerg]

noticed a minor additional difference... apparently I came to switch preq's for the pikes and halberdiers, but I don't suppose this makes any difference?