AI unable to build big army (>25000 pts)

[DRG]

Bottom line

When we increased the unit and formaton count years ago we considered what the vast majority of players would be comfortable pushing around and 500 units per side was considered to be way more than most would want to play with but it did allow for players to build really big games. It's inevitable that wherever limits are set someone will want to exceed that. Had we allowed 600 units somebody would want 700.

The problem with the AI bumping into the unit's ceiling wasn't an issue becasue the points limit was lower. We increased the points in WW2 to match MBT where the max points really is needed becasue of the higher tank costs.

So the best advise for the few players who are running into this problem is to keep your games within reasonable limits.

As interesting as 1000 unit assaults might be for a tiny % of the people who play it's simply not practical to tear the game apart to do it

Don

[Imp]

Whoa that totaly redifines my idea of a big game.
200 - 300 probably tops is a big game in my view, & yes the time to play one turn can take a while. Never even occured to me people would try & go larger & most people seem to think this is massive. 300 units seems to be able to reccreate any formation even the likes of modern Soviet US Tank Reg with 90 or 100 tanks a pop

[francoisD]

he guys,

i think you totally misunderstood my messages. i never ever asked to increase the total number of units per side in the game (500). i already think that 750 units (250 for me and 500 for the AI) is large enough and sometimes even become borrowing when i have to move several dozens of units in a turn...

i simply ask to make the Ai buy more expensive units if with its usual pattern of purchase, it will not come close to its total allowance of points (let say 40k pts) for the big battles...

i think that this is a sensible proposition that does not question all the previous hard works worth hundreds of thousands of hours ...

you can critize my request to be able to play with a german core worth 12000 pts and say its unreasonable, but focus on your critics.

i still wonder why it is possible to play with americans, british russians with armies worth 30000 pts with no problem, because they will finish as attackers, but not with the germans, because they will finish as defenders.

i do think this is a (minor) problem with the game.

[Mobhack]

Originally Posted by francoisD

he guys,

i think you totally misunderstood my messages. i never ever asked to increase the total number of units per side in the game (500). i already think that 750 units (250 for me and 500 for the AI) is large enough and sometimes even become borrowing when i have to move several dozens of units in a turn...

i simply ask to make the Ai buy more expensive units if with its usual pattern of purchase, it will not come close to its total allowance of points (let say 40k pts) for the big battles...

i think that this is a sensible proposition that does not question all the previous hard works worth hundreds of thousands of hours ...

you can critize my request to be able to play with a german core worth 12000 pts and say its unreasonable, but focus on your critics.

i still wonder why it is possible to play with americans, british russians with armies worth 30000 pts with no problem, because they will finish as attackers, but not with the germans, because they will finish as defenders.

i do think this is a (minor) problem with the game.

But I did try simply buying the 30 tank Russian heavy tank battalion every pass through the AI buy loop. It made little difference to the total buy points expended. About 32K points went up to about 35K points. The limit was the 500 units, as our experiments proved. With 1500 units a side, and a 650 formation limit then the WW2 Russians could actually buy 65K points.

IS-2 cost about 97 points. Even if you manually bought 499 of them, that would only be 48K points. Shermans cost about 80, fireflies about 100. The only other tank the allies have that can cost that much is Centurion and Pershing and very few arrived of those. A Panther costs nearly 130 points for 70 experience. A king Tiger costs 214. The German can run out of points fleet buying those!.

Most WW2 armies will only buy about 30-32K points with the normal AI mix of units (infantry, arty, and tanks).
(Easily checked - assign 65K max points, set player 1 as russian (or whoever you are interested in) and computer buy for an assault. In the deployment screen go to the HQ menu where the No. of formations, units and the points spent is displayed).

The problem is that the Germans have the ability to buy hugely expensive toys. The Allies have no access to 130 and 200 point tanks. So it is easy for a German player to unbalance things if he fleet buys expensive "cats". As Don said - just try to avoid building a very expensive core if you want the enemy to have a chance to meet it. The game is meant to be played with a historical battalion sized force, say a tank company and 2 of infantry with a few support troops.

Or simply play allied campaigns with the Nazis as the enemy - they will be in defence at the end in the main. Which is the best route for interesting games IMHO as then the AI has the force that is able to counter everything you can buy (provided you do not go T34/KV crazy in 1941 if playing Russian).

Cheers
Andy

[francoisD]

andy,

do you really think a 12000 pts core is a 'very expensive core'?

i do not.

i am not a crazy man asking to be able to play with 200 tigers 2
just a battalion of panthers and a battalion of infantry plus support...

i do think there is a problem with the game, because you can build much more expensive core (25000-30000) with the americans,russians and britain, with no problem at all, because the ai would buy armies worth 10000-15000 pts with the germans as defenders

perhaps a second button similar to the ai tank heavy on should be added, with off, nothing changes, with on, the ai avoids the cheapest tanks...

used and usefull only for the guys who want to play with germans and a size core of 12000 pts

well, i hope you will not consider me as a geek

[DRG]

Originally Posted by francoisD

andy,

do you really think a 12000 pts core is a 'very expensive core'?

i do not.

i am not a crazy man asking to be able to play with 200 tigers 2
just a battalion of panthers and a battalion of infantry plus support...

i do think there is a problem with the game, because you can build much more expensive core (25000-30000) with the americans,russians and britain, with no problem at all, because the ai would buy armies worth 10000-15000 pts with the germans as defenders

perhaps a second button similar to the ai tank heavy on should be added, with off, nothing changes, with on, the ai avoids the cheapest tanks...

used and usefull only for the guys who want to play with germans and a size core of 12000 pts

well, i hope you will not consider me as a geek

Maybe someone else can explain this so you can understand this but from where I sit both Andy and I have already done that.

The "allies" can build bigger cores becasue the Germans have more expensive "toys" that build up points faster without running into the units limit. The units limit is the "problem" in this case which is why we ran experiments with huge unit and formation limits to see where the boundaries actually are.

And NO we cannot code the game so "the ai avoids the cheapest tanks..." and as Andy already explained loading up the AI will allied heavy tank coys etc does not make much of a dent in the "problem" becasue the tanks themselves don't cost as much as the German ones do so don't run the points count up as quicky as it would for the Germans AND that would throw the balance of forces off with so many heavy tanks

Don

[francoisD]

Don

You have been the most severe with me,

Speaking of pigs with wings…is quite insulting. You say I do not understand anything, so that I am a feeble man…

I have always been polite in all my posts.

And contrarily to what you say, I understand well how the game works on this particular point, thanks to the numerous trials I have done.

I have identified a problem, at least for me, and asked for a solution. You could reply that you have many things to do before than solving this minor issue, and I would perfectly understand

But do not say that I understand nothing, or that it is impossible or too complex to do what I request

Because I think you are wrong and I am right.

Indeed, when you say that quote

becasue the tanks themselves don't cost as much as the German ones do so don't run the points count up as quicky as it would for the Germans

how can you explain that the germans, although they have the biggest tanks, are only the second or third armies in pts, behind Americans and british armies?

When you claim that you cannot control the points of an army, how can you explain that an army in exactly the same setting, can go from a 16000 pts army to a 26000 pts army?

If an army can go from 16000 pts to 26000 pts, that is exactly what I am saying: you can control the size of the army in points, but not with the game engine currently. Something has to be added.

This is the issue I raised.

Now I will explain what I have deduced from my experiments, and you will easily tell me if I am right or not

Point1

According to the date, type of battles, and so on… you determine what classes the AI has to buy
For example, the Ai has to buy 30% or so of tanks, roughly 45% of infantry, 15% of artillery, 10% of supports...

Then point 2, according to a rarity code ratio table you designed, the ai buys troops according to point 1. it does so until it exhausts all the points it could spend or until it reaches a size of 500 units.

This rarity code ratio table is simply acting like a probability distribution, and you use a random wheel like in a lottery to determine what troops are actually used…

That’s it.

So effectively, with this type of code, you absolutely cannot control the value in pts of the final army, and this explain the huge variance of the army points. And this also explains that to raise the pts of the army, you have only one degree of freedom: the maximum number of unit, 500...

Now I simply says that it is possible to raise the quality of the troops so that the Ai will build armies bigger in size.

Although in depth testing is required, you can simply define several rarity code ratio tables! Possibilities are endless: you have currently the rarity table for standard forces. You could create a table for elite forces, with expensive units (not only tanks, but artillery and so on). Probability degrees are simply the ones of the standard force, shifted to expensive units

But you could also design a rarity table for cheap support troops, with probabilities shifted to cheap units…

So that now with a 5000 pts battle, the game engine would have the choice to build a standard force, an elite force, and an army build of tips and bits of cheap and broken units

You simply now add a new probability distribution on top to choose what forces you can have, that depends on total number of points to spend. If the points to spend are higher than 10000, then choosing an elite force is compulsory.

This is so simple tha I wonder if this is not already the case

Now I will provide a small academical example with the cheapest army, the Russians, that proves clearly that within the absolute limit of 500 units per side you can build a very expensive army with the above proposition, at least 50% bigger than the typical current Russians army in January 1944 (average pts equal 19000).

70 kv 85, 70 t34 1943 equals 11875 pts for the tanks

32 hvy howitzer 152mm, 7500 pts

300 standard infantry worth each 16 pts, 4800 pts (infantry costs always the same in all armies, you do not need to rewrite the rarity table)

20 heavy at gun 61 pts each= 1200 pts

80 units worth each 40 pts, easy, means 3200 pts

That equals with a (standard) elite force of 500 unit, 29000 pts…

Now with this I can build a german core of 12000 pts

[DRG]

Well, what this all boils down to is this

1/ Both Andy and I have tried to explain this without success and seeing how this has been going I do not personally hold out any hope that doing it again will change the outcome

2/ We've been developing this game for 10 years so take it as a given we have a clue how it's all put together . Very rarely are we surprise by something.

but..

3/ You think we're wrong and have the solution

Despite what you may think both Andy and I have already invested quite a number of man hours building test EXE's and investigating this. Tearing the entire code that picks forces apart is not, and never will be, an option.


I can see what you..... "have deduced from my experiments" and what you have deduced about the way the AI picks appears to me to be misleading at best . The AI picklist in winSPWW2 for the Russians alone is a very long and very complex set of commands that runs 1,159 lines of code and is not simply "30% or so of tanks, roughly 45% of infantry, 15% of artillery, 10% of supports... ". though that is a simplified explanation of the end result. The entire set of picklists is a massive amount of code designed to produce a balanced force. What you propose unbalances that and undos hundreds of hours of work simply to get a high points count on the map


Then we get into language difficulties. I'm not quite sure what to make of......"according to a rarity code ratio table you designed". See..... I have no idea if you think we have designed and use a rarity table or that we should. The rarity part of the radio code is there mainly to ensure rare, low production vehicles are not bought too often and common ones are and has NOTHING to do with how much they cost.



Don

[RERomine]

If I may, there seem to be two conflicting points being brought up:

1. The Germans have more expensive equipment.
2. The Germans spend less on an AI selected force than the Americans, Russians and British.

The majority of the tanks they buy are medium tanks and the most common one purchase is the PzKw IVj(+), an 89 point unit at 70 experience for a 4/1945 battle. They also seem to like the PzKw IVh (88pts). The Russians lean heavily on the T-34/85 M1944 model (92pts), the Americans on the M4A1(76) (101pts) or the M4A3(76) (102pts) and the British, well there are a host of them including the Sherman III (72pts), Sherman V (81pts) or the Comet (89pts). When the British have Shermans, each platoon seems to have a Firefly as a gun tank and this pushes the average per medium tank purchased. That explains stuff like below:

1945 ON A (30722, 32255) G(26318, 28074), R(26768, 28398), B(27258, 28793)

Since the Americans have the most expensive common medium tank, it explains why their typical AI selected core force is higher. Once in a blue moon, the German go heavy on Panther Gs and that skews those numbers all to heck.

As far as heavy units, they tend to buy no more than 12 and they seem to be nine Tigers and three PzKw VIbs. This type of heavy tank purchasing seems to be typical of all countries. What I haven't been able to get the Germans to do is to buy out of the JagdPanther Group, meaning no JagdPanthers or JagdTigers, two of those expensive toys. They don't get Hetzers, Marders or JPz IV/70s, either. I do see Stug IIIgs, but those are coming out of the StuG Group, because they are defined as "Assault Guns" and not "Tank Destroyers". The Americans, Russians and British all purchase tank destroyer type units. One would think the Germans should by some tank destroyers as they were cheaper to produce than tanks. I've tried a dozen times and haven't been able to get them to purchase any allotting the AI 50000 points to use.

I don't know how the force selection code loop works, but if the JagdPanther Group is being skipped or is not included, it would likely cause other units to fill out the 500 unit limit. The vast proportion of those units are much cheaper and it would consequently push down the value of the total core. Out of all this, I'm trying to provide something you can sink your teeth into. Without seeing the code, all I can do is look at output and speculate, but if I am correct it would be a bug.

As far as purchasing large numbers of heavy tanks, I'm not going to step into that one because it seems to be more a factor of personal taste. Seems to me if someone wants lots of heavy tanks, they could go into the OOB (after saving the vanilla version) and start chopping out units they don't want used. If you limit the choices the selection loop has, it will have no other option but to select heavier stuff.

[DRG]

Originally Posted by RERomine

The majority of the tanks they buy are medium tanks and the most common one purchase is the PzKw IVj(+), an 89 point unit at 70 experience for a 4/1945 battle. They also seem to like the PzKw IVh (88pts).



Once in a blue moon, the German go heavy on Panther Gs and that skews those numbers all to heck.

It's interesting you would mention this. We had noticed that the AI as the Germans picked Panthers, as you noted, only "once in a blue moon" which is far less than it should have and the V3.5 OOB's have been changed so Panthers will be picked more often from April 1944 onwards. There won't be a flood of them but it will be more likely in V3.5 than previously and all of that was a direct result of how the code reads through an OOB and sorts the rarity part of the radio code. The problem is, and this is fundamental to the way the code works and we've adapted the picklists and OOB's themselves to accommodate this, is it reads units from the last number and works it's way UP. In smaller OOB's with fewer units to pick from this isn't usually a problem but the German OOB is somewhat more complex than the rest and what was happening by the time you got to the summer of 1944 was there were a large number of tanks for the AI to choose from before it got to the first Panther that was available. You cannot simply take a vehicle from one slot and put it in another or substitute units because save games and scenarios record slot number when they save. Switch a Panther for a PzIII and you have real problems so we added a new Panther near the end of the list and the AI is happy to pick it more often

Originally Posted by RERomine

What I haven't been able to get the Germans to do is to buy out of the JagdPanther Group, meaning no JagdPanthers or JagdTigers, two of those expensive toys. They don't get Hetzers, Marders or JPz IV/70s, either. I do see Stug IIIgs, but those are coming out of the StuG Group, because they are defined as "Assault Guns" and not "Tank Destroyers".

The AI will buy formation 44 (Jagdpanzer Grp) or 16 (JagdPz Zug/4) if a random number greater than 60 is generated with 99 being the max so call that a 40 % chance you might see them from July 1943 onwards IF the AI goes down that thread of the picklist looking for things to buy ( and theres another random chance that might happen ). A Jagdpanther has a normal chance of being picked but is unit 53 of 999 and refer to what I told you above as to why they are rarely seen ( in the game AND in reality ) but there is a chance that maybe, someday, it might pick one but ONLY in a delay OR defend. They are not set up to be bought in any other battle type because delaying and defending is their purpose --not advancing so yes, Jagdpanzers ARE part of the picklist but they're not going to pop up very often ( and that doesn't make it a "bug" )


Don