|
|
|
Notices |
Do you own this game? Write a review and let others know how you like it.
|
|
|
December 9th, 2008, 09:46 PM
|
|
Shrapnel Fanatic
|
|
Join Date: Mar 2005
Location: GWN
Posts: 12,488
Thanks: 3,957
Thanked 5,691 Times in 2,811 Posts
|
|
Re: AI unable to build big army (>25000 pts)
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
|
December 10th, 2008, 03:08 AM
|
Captain
|
|
Join Date: Oct 2008
Location: Columbus, Ohio
Posts: 975
Thanks: 1
Thanked 14 Times in 12 Posts
|
|
Re: AI unable to build big army (>25000 pts)
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:
Quote:
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.
|
December 10th, 2008, 05:35 AM
|
Corporal
|
|
Join Date: Dec 2007
Location: France, Lille
Posts: 76
Thanks: 2
Thanked 0 Times in 0 Posts
|
|
Re: AI unable to build big army (>25000 pts)
don,
first thx for answering, because that proves that, at least, you read my messages.
quote:
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.
this is exactly what I proposed: use a second rarity table to shift probabilities to expensive units if the points to spend are higher than 10000 or 15000. similar to the ai heavy tank button,
but i do not like numerous tanks on the battlefield, and i prefer to use the button off.
however i do like to play with big battles, and everything is already present in the game, except on one very minor point, because the ai never build big armies.
i know you have worked much more time than me on this game. in fact i am simply a player and YOU are the developper. I respect you.
However you should consider that I have also tested the game during many hours, on this particular point. Why have I done this? it is useless for a player that simply wants to win at all costs and that will buy 200 tigers 1 anyway... (with respects to the realism discussion mentionned in my first message)
i want to play a balanced game, but funny as well, since this is a game.
It seems to me that the game engine could perfectly hand battles with 25000 guaranteed pts per side within the limit of 500 units,
well, you know my position, but you are the guy who works and decides. I will respect your decision whatever it may be.
if you do not improve this game as i proposed, well, i have two solutions: 1) forget the germans,
2) play with a single compagny of panthers. but i fear that even in this game, with a 9000 pts core, the game could very well be unbalanced. indeed with the huge variance I have noticed, it could happen very well that the russians buy armies worth less than 22000 pts (sure with heavy off, possible with heavy on).
in fact i am already facing this problem now in afrika, 1941, with a 5500 core in delays battles. I am not sure if the ai has an army worth 12500 pts. in the last two battles, delays, i checked at the end of the battle the number of units the ai had to be sure it was less than 500, a proof that it exhausted all the points it had. and in fact in one battle i have reached 500 units (i count the number of pages, because there are 25 units per pages, if there are 20 full pages, the number of units is 500). what i did? i reloaed the save at the end of my last battle because this one was unbalanced.
so i am asking to you and andy a very very small improvement of the game.
since the ai might build armies strong enough, could you add to the deploy screen of the human player an information about the pts of the ai army and how much it should have?
so if human players see that the ai does not have a correct army, he restarts a new battle from the last save, instead of playing a full battle just to see at the end that the battle was unbalanced...
|
December 10th, 2008, 05:42 AM
|
Corporal
|
|
Join Date: Dec 2007
Location: France, Lille
Posts: 76
Thanks: 2
Thanked 0 Times in 0 Posts
|
|
Re: AI unable to build big army (>25000 pts)
thx to reromine too for his experiments that confirm that the rarity table could be modified.
i am not sure if i will modify the oob files, i have never done this before. anyway thx for the tip
|
December 10th, 2008, 05:51 AM
|
Corporal
|
|
Join Date: Dec 2007
Location: France, Lille
Posts: 76
Thanks: 2
Thanked 0 Times in 0 Posts
|
|
Re: AI unable to build big army (>25000 pts)
i have just found out another soution to the problem of big battles:
the ai purchase loop should be redone until the total number of units of the ai is below 500. indeed 500 is a proof that the ai has not enough points.
this is a very very very small modification of the code, however, it raises strongly the duration of the purchase step, because you rely solely on probabilities...
and it might never stop...
|
December 10th, 2008, 10:21 AM
|
|
Shrapnel Fanatic
|
|
Join Date: Mar 2005
Location: GWN
Posts: 12,488
Thanks: 3,957
Thanked 5,691 Times in 2,811 Posts
|
|
Re: AI unable to build big army (>25000 pts)
Quote:
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
Quote:
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
|
December 10th, 2008, 11:27 AM
|
|
Shrapnel Fanatic
|
|
Join Date: Mar 2005
Location: GWN
Posts: 12,488
Thanks: 3,957
Thanked 5,691 Times in 2,811 Posts
|
|
Re: AI unable to build big army (>25000 pts)
Quote:
Originally Posted by francoisD
don,
first thx for answering, because that proves that, at least, you read my messages.
|
Yes, I read your messages and as I have already explained both Andy and I have expended a number of man hours looking into this issue. However. It's increasingly clear that we are going around in circles and you do not want to hear what we have to say so, on this thread the time for reading what you think is rapidly drawing to a close because François all this " You think we are wrong no matter what we tell you " stuff is turning into an enormous waste of my time
Quote:
Originally Posted by francoisD
**************************quote DRG*********************************************** ********************************************
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.
************************************************** ************************************************** ******************************
this is exactly what I proposed: use a second rarity table to shift probabilities to expensive units if the points to spend are higher than 10000 or 15000. similar to the ai heavy tank button,
|
OK, lets stop right here and get this straight once and for all. Perhaps this is a language issue
You say.......... "use a second rarity table "
Where exactly do you think the FIRST " rarity table" is ???. Explain this without typing an essay and we'll go from there.
As I have explained the rarity of a vehicle is its commonality as part of the radio code x0 would be normal, x1 would be rare, x2 would be more common. The code reads that when it's picking units from an OOB. You seem to think it would be "easy" to add a bit of code that reads cost that would build a force then, if the unit limit was reached would build another force looking for higher cost units and if that still bumped into the unit limit would run through the picklist AGAIN looking for even more expensive units to buy eventually selecting a very "elite" force when it needs one to deal with the unit limit issue.
And I tell you, if this is indeed what you propose, that despite what you may think there is NOTHING "easy" about any of that.
Quote:
Originally Posted by francoisD
but i do not like numerous tanks on the battlefield, and i prefer to use the button off.
however i do like to play with big battles, and everything is already present in the game, except on one very minor point, because the ai never build big armies.
i know you have worked much more time than me on this game. in fact i am simply a player and YOU are the developper. I respect you.
However you should consider that I have also tested the game during many hours, on this particular point. Why have I done this? it is useless for a player that simply wants to win at all costs and that will buy 200 tigers 1 anyway... (with respects to the realism discussion mentionned in my first message)
i want to play a balanced game, but funny as well, since this is a game.
It seems to me that the game engine could perfectly hand battles with 25000 guaranteed pts per side within the limit of 500 units,
|
Sure you can, I could build an under 400 unit , 25,000 point Russian force quite easily by loading up with an unreasonable amounts of artillery and Heavy tanks . But you tell me you play with Tank Heavy preference OFF
I'll tell you what........ YOU set up a game , use whatever points you think it will take to try and make your point then save it BEFORE "quite deploy" so we can see the formation, unit and points details, then post it here and we'll all see what you would want the AI to buy to build a "balanced force" that....
1] isn't tank heavy
2]is as complex as what the AI buys now with the same unit variety
3]does not buy wildly non historic types of vehicles ( so no fleets of JadgTigers or ISIII's or nothing but dozens of 88 AT guns )
......that uses all the points you think it should. I will be most interested in what you show us all
Don
|
December 10th, 2008, 11:38 AM
|
Captain
|
|
Join Date: Oct 2008
Location: Columbus, Ohio
Posts: 975
Thanks: 1
Thanked 14 Times in 12 Posts
|
|
Re: AI unable to build big army (>25000 pts)
Quote:
Originally Posted by DRG
Quote:
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
|
Hypothetically, to play a campaign with lots of heavy tanks, you could copy the original OOB and modify it for that purpose as long as you restored the the original before you try to load old saved games and scenarios, correct? Basically, one OOB is used for a specialized campaign and the original OOB for all of the out of the box scenarios.
Quote:
Quote:
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
|
That makes sense as they JagdPanther and JagdTiger were too slow for any real offensive actions. When I was poking around, it was set up as a "Meeting Engagement" and sounds like it would exclude those. What about smaller tank destroyers? I didn't see Hetzers, Marders, JPz IV/70s or other turretless tank destroyers either. As I mentioned, the StuG IIIg was selected, but off of the assault gun list. Does the selection code considers German assault guns to be interchangable with tank destroyers? The Russians and British seem to have their turretless tank destroyers show up frequently. I don't think the Americans had any if you don't count those half-track mounted guns.
|
December 10th, 2008, 12:08 PM
|
Corporal
|
|
Join Date: Dec 2007
Location: France, Lille
Posts: 76
Thanks: 2
Thanked 0 Times in 0 Posts
|
|
Re: AI unable to build big army (>25000 pts)
hi don
indeed there was a problem of understanding what you said. this is clear now, thanks to roremine, who told me to look at the oob files
i now understand what is the radio code, i looked before your answer to the mobhack help file.
i will tell you now what i believed first:
there was an internal table that gave each unit a probability to be chosen, within each unit classes (so that each probability added to one withing the same class), while there is another probability on top for the classes themselves. i also believed there was a pre selection of troops because whithin the same classes, not all the troops are chosen within the same battle.
so I said to shift the probabilities within classes to expensive units
but this is not the case
now there is only 3 cases for each unit, normal, rare, and common, and from this, you compute probabilities by another mechanism
so we cannot directly handle the probabilities.
i was looking at the oob files for the first time of my life when i read your message. apparently, i could create a new oob set that says that all expensive units are common, and all others are rare. do not know what will this give.
now for the example your request, it was already given in one previous message.
i think it is similar to heavy off, because there is only 30% of tanks (with heavy on, its 50% or so)
janurary 1944, russians, meeting engagment, 29000 pts
300 standard infantry worth each 16 pts, 4800 pts (60% of infantry)
70 kv 85, 70 t34 1943 equals 11875 pts for the tanks (28% of tanks)
32 hvy howitzer 152mm, 7500 pts (6% of heavy artillery)
20 heavy at gun 61 pts each= 1200 pts (4% of heavy AT gun)
78 units worth each 40 pts (flak, FOO, ...), easy, means 3100 pts
i do not think this army is so unrealist. it is an elite force...
so it is clear that, with heavy off, you only have 200 units to adjust the pts for the big battles, because infantry always cost more or less the same thing
anyway, you can critize this army
but do not you think that the huge variance in pts encoutered with your algorithm is a drawback?
you set up a 20000 pts battle and you end up with the ai having a 16000 pts army?
i do not think the variety in troops should reflect a variance in pts, especially when considering the points can never overshoot the total pts.
if it could, you could say that the game engine is purposely having a huge variance so that battles can be randomly easy, difficult, and so on..., independantly of anything else
and what about the idea of giving on the deploy screen the total pts of the ai army and the total pts it should use?
for me the problem would be almost solved in this case, i just would have to reload and restart battles, until it gives me a correct ai army.
thanks to the huge variance, since your game can provide sometimes, 28000 pts armies
a bit repetitive...
|
December 10th, 2008, 12:19 PM
|
|
National Security Advisor
|
|
Join Date: Mar 2005
Location: Dundee
Posts: 5,955
Thanks: 464
Thanked 1,896 Times in 1,234 Posts
|
|
Re: AI unable to build big army (>25000 pts)
Quote:
That makes sense as they JagdPanther and JagdTiger were too slow for any real offensive actions. When I was poking around, it was set up as a "Meeting Engagement" and sounds like it would exclude those. What about smaller tank destroyers? I didn't see Hetzers, Marders, JPz IV/70s or other turretless tank destroyers either. As I mentioned, the StuG IIIg was selected, but off of the assault gun list. Does the selection code considers German assault guns to be interchangable with tank destroyers? The Russians and British seem to have their turretless tank destroyers show up frequently. I don't think the Americans had any if you don't count those half-track mounted guns.
|
Examine the German formations.
Some are Assault guns - those get bought to support in any battle. The units available in those units will be bought randomly according to radio code.
Some are tank destroyers, and those will be bought usually in the defence only. Units bought will be a rnadom pick as determined by radio code from those available to those formations. (I have just destroyed some JPz 1 in a meeting engagement in the western desert with my valentines, so it will buy SP-AT formations in meeters as well)
There is a formation that uses 2 or 3 marders exclusively as far as I recollect - I think that one gets bought on low points remaining.
Stugs in the German army can appear as tanks, assault guns, and even as tank destroyers at various times.
The German AI pick is a very complex piece of code, and date, battle type, points remaining, snow, city, batloc, who the actual opponent is and several other factors in the 2-3K lines of code determine what gets bought. So simply running a single purchase trial is only a guide. Try 50-100 or so of each battle type at a specific batloc and points value, then analyse the trends.
(e.g if the USA want to see more tigers than the average - try the Bulge time frame USA as opfor to AI purchase Ger. AI skijaegers - try a snow map. More ger AI paras at cassino maps)
Cheers
Andy
|
Posting Rules
|
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
HTML code is On
|
|
|
|
|