(1.3.3)Is new method for resource shortages the underlying cause of German AI problems?

[Meglok]

@Dalwin @Gwydion5 @Pro_Consul and anyone else following this thread.

Let me elaborate, I was perhaps imprecise in my wording earlier. I think the issue in 1.3.3 (and earlier) is an interlocking stew of issues, in that one problem could be an contributing part of another problem.

Example-the front management shuffle. Based on what I have seen and read from developer's posts, the ai nation creates fronts based upon a need to attack from, defend on, or garrison along a section of it's national border. The ai tries to judge how many divisions it needs for each front and the priority of that need and will then either grab existing divisions or create new ones to fill the need. When the ai begins to shuffle units, it is either responding to a division call from another front or a perceived weakness in it's assigned front. This is without a doubt one of the biggest issues out of modder's control and the cause of many game disasters. But why, and what are the underlying causes?

The basic problem comes in when there are not enough divisions to meet all the needs. @podcat was very clear when he said the battle planner ai functions better with more divisions. Obvious statement, if the ai has enough tools it can fill all holes and there is no shuffle. I know when I go on a tag frenzy in 1.3.2 and manually gave ai Germany all of it's minor's divisions, it will function quite well. Same with Italy, pre 1.3.3, now in 1.3.3 it just continues to send divisions to Africa.

The base problem is what the ai does when it doesn't have enough divisions. The ai apparently doesn't perform well in allowing existing need where a division is currently assigned to override a call to reinforce elsewhere. It yanks divisions from point A to fill a perceived hole at Point B, which opens a hole at Point A, which it will within 1-2 weeks grab divisions from Point C to fill Point A, and so on.

Prior to 1.3.3, a LOT of players complained about the ai producing units that it could not equip, the division spam issue. Part of this was the expeditionary forces issue, the ai would send off all it's divisions as ExpForces, and then keep creating more divisions as long as manpower existed irregardless of ability to equip or resupply existing divisions. This led to ai nations with under-equipped toothpicks and massive lag. So now, there looks to be a tightening down of ExpForces being offered to ai nations. This of course impacts the front shuffle issue adversely, as the ai doesn't have as many tools in it's tool box to play with.

PDS finally fixed the convoy bug, but that has impacted the African safari issue by allowing the ai to sends tons of divisions to the Congo, reopened the Axis suicidal conga line of transports thru the English Channel again, AND restarted the constant micro invasions by the Allies of occupied Europe the ai Axis can't deal with. All of these also impact the battle planner shuffle issue, because if Canada just landed 10 divisions at Brest, or, Italy just sent 30 divisions to Italian Somaliland, the ai has to pull defenders from somewhere..... When the ai is doing these at least once a month, the ai has no prayer of functioning properly.

The new industrial, infrastructure, and resource changes look to be a change too far. I think I can understand what they were trying to do but the ai can not cope with it. Inf Eq is prioritized over all to help prevent under-equipped divisions, which leads to too many other production lines being penalized or shut down. The ai will not trade overseas for resources now (to protect convoys from going thru the Channel??), which just cripples ai Japan. No one will trade with her. This of course leads to fewer divisions, which leads back to division shuffle.

For some unknown reason, on top of the ExpForces issue ai allies and minors in 1.3.3 will sometimes not even leave their borders. If Hungary and or Romania stay at home Germany is in trouble in the East. And guess what happens, the unit shuffle dance.

Division templates have been upgraded again, but still suffer from a one size fits all generic code that does not reflect an ai nation's resource availability, industrial capacity, infrastructure, or CIC capacity. Example, the standard generic default infantry template is now 7-INF/2-ART with support of eng, rcn, hosp, AT, and Art. Now while the ai won't (to my knowledge) create that template until it has all the tech, how is any minor nation, even Commonwealth nation or Axis minor supposed to build all of that equipment under the new production rules? What I am seeing is the ai will build the Inf Eq, but will usually deploy divisions without ART, AT, and or support equipment. If it deploys the divisions at all. Which either leads to toothpick divisions or a lack of divisions, circling back to causing the division shuffle dance. National templates are a must for most nations that reflect the new reality.

Africa and the never sufficiently cursed and damned Congo shuffle issue is back. PDS refuses to block off the Sahara from sub-continent Africa and now that the ai has convoys it pours divisions into Africa. I understand the ai sees a front line and wants to either attack across it or defend it's territory. I understand that for some reason on this issue PDS insists on historical accuracy. But the damn issue is contributing to breaking the game. Too many divisions are heading south and it is screwing Italy. Divisions in sub-Sahara Africa can not respond to an invasion of Sicily! Please PDS , for the love of all that is holy, adopt No Man's Land or some derivative and block off sub-Sahara Africa, or make these territories colonies/puppets so that the mother country won't pour divisions into them (see Fox and Lion mod). These changes sure seems to help the game run better according to everyone who uses them, and isn't that the goal, a game that runs better and is fun to play?

I suspect that if PDS is able to tackle, resolve, or tweak many of the underlying issues the battle planner shuffle will calm down. They just have to make fixes now in a hot fix and not wait until 1.4. The single player game is almost unplayable now unless you always play Germany. And that gets very boring very fast.

 

[Dalwin]

I do like your stew metaphor, but considering the inter-related nature of how various factors affect each other, I prefer to think of it as a web.

Lower production --> fewer units --> more problem with front assignment and BP shuffle

fewer units --> less able to perform suppression --> lower production

lower production --> fewer convoys --> harder to import resources --> even lower production

not enough units to cover front --> more BP shuffle calls to plug holes --> even fewer units actually at the front

Even from that very abbreviated causal effect list one can see that not only are these things inter-related, but that many of them form feedback loops. Below a certain level of something critical like a key resource or number of units, the situation gets locked into a death spiral and collapses completely, often quickly.

 

[Pro_Consul]

The basic problem comes in when there are not enough divisions to meet all the needs.

I don't really disagree with anything you're saying. But I would say my own point of departure as far as where my position begins to diverge from begins with that sentence above. And your more detailed explanation reveals how I have been consistently failing to properly explain my own position. The shuffle, the self-reinforcing nature, the fact that unit count can induce and/or increase this behavior...all true. But with AI Germany in particular the really game-breaking part, the persistent meltdown on the Eastern Front every single game, seems to begin with one simple issue: it is failing to properly determine what its needs actually are.

Look again at the cases that have been posted in the last week or so. What is the common denominator in the German meltdowns on the Eastern Front? Africa. For some inexplicable reason, the German AI is suddenly calculating a need for dozens of divisions in Africa. And it doesn't do that until after France falls, right before or as it is launching Barbarossa. That phantom need is, in my own opinion, the single largest element in the current chronic meltdown issue. Even when the human player manages Germany's production, ensuring it has adequate forces to meet its actual needs, when the player turns that over to the AI, the AI still self-destructs because of its insistence on committing large portions of its total force to phantom needs. Yugoslavia also tends to be such a phantom need, though one that seems not to suck in near as many divisions.

It is my prediction, though one I hope we never have a chance to test, that if the phantom need problem, particularly in Africa but also in Yugoslavia, were fixed, and everything else remained as-is (the part I don't want to test), that AI Germany would suddenly begin to act in a more or less sane manner during Barbarossa. The production and other issues would almost certainly result in no real chance of winning, but at least it wouldn't be such an instant and spectacular self-destruct sequence every single time. Or so I think anyway.

 

[Dalwin]

I don't really disagree with anything you're saying. But I would say my own point of departure as far as where my position begins to diverge from begins with that sentence above. And your more detailed explanation reveals how I have been consistently failing to properly explain my own position. The shuffle, the self-reinforcing nature, the fact that unit count can induce and/or increase this behavior...all true. But with AI Germany in particular the really game-breaking part, the persistent meltdown on the Eastern Front every single game, seems to begin with one simple issue: it is failing to properly determine what its needs actually are.

Look again at the cases that have been posted in the last week or so. What is the common denominator in the German meltdowns on the Eastern Front? Africa. For some inexplicable reason, the German AI is suddenly calculating a need for dozens of divisions in Africa. And it doesn't do that until after France falls, right before or as it is launching Barbarossa. That phantom need is, in my own opinion, the single largest element in the current chronic meltdown issue. Even when the human player manages Germany's production, ensuring it has adequate forces to meet its actual needs, when the player turns that over to the AI, the AI still self-destructs because of its insistence on committing large portions of its total force to phantom needs. Yugoslavia also tends to be such a phantom need, though one that seems not to suck in near as many divisions.

It is my prediction, though one I hope we never have a chance to test, that if the phantom need problem, particularly in Africa but also in Yugoslavia, were fixed, and everything else remained as-is (the part I don't want to test), that AI Germany would suddenly begin to act in a more or less sane manner during Barbarossa. The production and other issues would almost certainly result in no real chance of winning, but at least it wouldn't be such an instant and spectacular self-destruct sequence every single time. Or so I think anyway.

I see no mystery at all in why Germany puts Africa as a high priority immediately after the fall of France. It is clearly a case of the AI's persistent single-mindedness in marching toward the sound of the guns.

Between the time when France falls and before the invasion of Russia, Germany shares no land borders with any enemies. One of their faction members does, however. The Italian borders with Egypt and Tunisia are the only active fronts in that sense. So to the AI's short sighted way of looking at things that front is top priority.

 

[Meglok]

I see no mystery at all in why Germany puts Africa as a high priority immediately after the fall of France. It is clearly a case of the AI's persistent single-mindedness in marching toward the sound of the guns.

Between the time when France falls and before the invasion of Russia, Germany shares no land borders with any enemies. One of their faction members does, however. The Italian borders with Egypt and Tunisia are the only active fronts in that sense. So to the AI's short sighted way of looking at things that front is top priority.

Yep, Germany needs to immediately progress towards Yugoslavia and Greece and North Africa, then Russia. Unfortunately this does nothing regarding Italy sending so many troops to Africa it has issues defending Sicily.

 

[Pro_Consul]

I see no mystery at all in why Germany puts Africa as a high priority immediately after the fall of France. It is clearly a case of the AI's persistent single-mindedness in marching toward the sound of the guns.

Between the time when France falls and before the invasion of Russia, Germany shares no land borders with any enemies. One of their faction members does, however. The Italian borders with Egypt and Tunisia are the only active fronts in that sense. So to the AI's short sighted way of looking at things that front is top priority.

Not to be confrontational, but I am not buying it. In too many cases, including the ones posted in the other active thread about this, Germany has absolutely zero interest in Africa until hostilities begin with USSR. Then suddenly half the Wehrmacht is urgently required in sub-Saharan Africa post haste. That is obviously not an issue of having no land borders with an enemy. There is something very, very hinky going on with its front calculations the moment USSR gets involved.

 

[Xia]

I'll be honest, I've never seen substantial, or even meaningful numbers of German troops in Africa; the Soviets normally beat the overwhelming bulk of the Wehrmacht on their own, and the reason is purely production related. I've seen the damned Soviets reach Madrid in 1942 whilst the Allies are still bottled in Brittany. The Soviets have more of everything, and 99% of the time in a fight between two AI nations, the attacker takes overwhelmingly lopsided losses - I was playing a peaceful Germany for an experiment on AI behaviour; in it the UK invaded Norway in ~1944, Norway and my 7 volunteer forces inflicted well over 1m losses on the allies for 0 land lost and ~50k troops lost. This can be seen all over the place when players are involved in AI vs AI conflicts.

It is COMMON to see Soviets lose 2-3m men for 4m German casualties before 1943 even, it is common to see the Germans bleed their divisions to death on the Maginot, in the swamps of the Benelux... I don't think I've ever seen an AI Germany achieve a positive Kill/Death ratio - despite Germany being the single major most dependent on good Kill/Death ratio. Even before 1.3.3 this was an issue - the Germans would lose 800k-1.5m men taking Poland and France for ~300k allied casualties - but back then they had the industrial base to recover their crippling losses, now? Nothing. Once those initial pre-war stockpiles are gone, AI Germany will never even vaguely equip its army again. It's a combination of lack of rubber (as all major rubber sources are Allied from the outset of the war), limited supply of oil unsafe supply of tungsten and low quantity availability of aluminium (no where near on the scale of shortages of other resources though).

It's not uncommon to see the Luftwaffe defeated in 1939 and the last Panzers destroyed in 1940 for example.

Does the AI obsessively marching over the Sahara help? No. But the small number of African divisions which are often posted as examples of the German AI's failure to prioritize wouldn't actually solve the biggest issue - 135 divisions vs 300 divisions is still a massively overwhelming German defeat.

 

[Meglok]

@Xia aluminum should not be a big issue for Germany. Iirc Hungary has a decent amount and it's puppet Vichy has plenty it can demand. Unless there are additional issues with trade and the AI not acquiring those resources.
Rubber is going to be an issue, the new overseas trade rules will limit Germany's desire to ask Siam and Japan and the AI doesn't seem to prioritize synthetic construction enough to meet needs. The new penalty to desire for overseas trade routes is too harsh to the Axis. If Germany is building trucks it's airplane production is in trouble.

 

[Dalwin]

I decided to observe the situation more closely by playing Hungary. I don't like hands off games and using the human AI command. This also allowed me to test how things would go if Hungary contributes somewhat. However, since I didn't spam garbage, January '41 saw me with 45 divisions instead of the usual 100+.

Things went well at first, and I wouldn't be too harsh in criticism of the Germans. Yes, they declared against Benelux before Poland fell. Yes, they did that while a large force of British and French had come ashore from Bremen to Rostock. The Germans coped with it fairly well. My only assistance to them in Poland was 200 fighters. I did send a dozen divisions to assist with the northern landings but I did very little of the work to drive those units into the sea.

The only help I gave in the conquest of France, Benelux, Denmark (in that order since the French had left gaps in the Maginot) was to garrison a few key locations after their capture. I had sent ~20 divisions for that purpose. I also had an Army of 24 in Hungary with 1/3 on Russian border and 2/3 on Yugoslav.

The Germans did a fine job (at first) of garrisoning all this and even with suppression. My forces were not needed right away. They did send units to Africa but I never saw more than 8 or 10 German units there at any one time, though I think some did cross the Med back and forth. The Italians were doing well until the Dutch took Tobruk and cut off a large force that had pushed into Egypt. Spain had joined the Axis and Gibraltar was closed.

The Germans and Italians sent a large force into southern Norway. This went well at first until for no reason all of the Italians and half the Germans left Norway (Bergen and Trondheim had not even fallen yet.) They all sailed en-masse through the Channel to various points in Western France or Spain or Spanish Morocco. Obviously a safe crossing of the Baltic followed by train rides to most of those locations or to Southern Italy or Spain first if crossing to Africa would have been much better. My air force tried to help these get through the Channel but many died.

This foolish loss of units started the downward slide. I was pleased to see that as soon as Germany began the War with Russia NF they transferred the bulk of their forces to the front to prepare for the war. Gone are days of when the first day of an AI declared war had nearly zero units on the front. They sent 175 divisions for this preparation which left a skeleton force in the west. If not for my Hungarian units there, several ports would have been empty. As it was, suppression went from completely under control to somewhat problematic. If not for my garrisons this would have been much worse.

Of the forces that Germany did leave in the west or in Norway, far too many are panzers. They do have panzers for Russia as well, but I think it should have gotten a higher proportion of them. I have not yet played past the point where Russia gets invaded. Rumania has not joined the Axis. Yugoslavia is still neutral and the Italians have only made slight progress in Greece.

The invasion will take place in January of '41. I think it would be much better if Germany delayed Barbarossa by a few months and helped clear out the Balkans first. The Italians and Spanish will probably lose North Africa but it is not definite yet. No Mans Land mod is in effect.

I will come back with further details later.

 

[Dalwin]

Rumania did join the Axis less than a week prior to the NF for Barbarossa being finished, in response to the annexation of Bessarabia. One odd thing I noticed is that the density of German troop deployments was twice as thick along the Hungarian and Rumanian frontiers in spite of those sections also having minor country troops present.

I also have to correct one point from the previous post. In spite of the War with Russia NF concluding in late January of '41, Germany did not actually declare war until March 7. I am not sure I have ever before caught the AI not declaring as soon as able even when doing so was a blunder.

In the midst of Germany struggling overall in 1.3.3 there are some key points whre it shows obvious improvement. (Well done @SteelVolt ).

 

[Meglok]

Rumania did join the Axis less than a week prior to the NF for Barbarossa being finished, in response to the annexation of Bessarabia. One odd thing I noticed is that the density of German troop deployments was twice as thick along the Hungarian and Rumanian frontiers in spite of those sections also having minor country troops present.

I also have to correct one point from the previous post. In spite of the War with Russia NF concluding in late January of '41, Germany did not actually declare war until March 7. I am not sure I have ever before caught the AI not declaring as soon as able even when doing so was a blunder.

In the midst of Germany struggling overall in 1.3.3 there are some key points whre it shows obvious improvement. (Well done @SteelVolt ).

I have been thinking, is it time to reset the Around Maginot and Weseruburg focuses back to where they started? Put Weseruburg before Around Maginot and change the lead time back to 70 days for AM, or at least longer than 28 days so that Germany is done with Poland before it turns West?

 

[Dalwin]

I have been thinking, is it time to reset the Around Maginot and Weseruburg focuses back to where they started? Put Weseruburg before Around Maginot and change the lead time back to 70 days for AM, or at least longer than 28 days so that Germany is done with Poland before it turns West?

That would help a little, but the AI got through all those timing difficulties in good shape. The way it played was stupid but it got away with it. In fact the bigger blunder was the Allied landing in Germany within days of Poland being invaded. That cost them 40 divisions or so and greatly hastened the fall of France.

Changing the timing back might actually make things worse. If Weserubung had come sooner and Denmark had consequently been in the war before the fall of Poland many of those British units would have been able to fall back instead of being trapped against the neutral Danish frontier.

 

[Dalwin]

Six months into Barbarossa and here is a quick summary.

The Hungarian sector, in which I started with only 8 divisions but increased it to 10, was at the gates of Kiev. My units were a minority there. This was also the area where Germany had deployed a large portion of its units.

The Rumanian sector to the south had kept pace and was threatening Dnepropetrovsk.

The German sector in the center had kept pace and was only a province or two behind the Hungarian sector, so not quite to the river yet.

In the North, however. Germany had quickly lost East Prussia. They went on to lose all of northern Poland. The front had moved into Germany and was threatening Stettin.

I could probably play more dynamically and shift my remaining 15 divisions from the neutral Yugoslav border to around Warsaw and drive north toward Konigsberg. That might save the war, but that is not what I am testing here. We already know that dynamic and well placed moves by a human player will put the AI to shame.

------------------------------

Summary: Germany did very well, more or less all by themselves, until a few weeks after they invaded Norway. At that point Germany, Italy and Spain all threw several units away in the Channel which resulted in troop shortages later. On the Russian front, Germany put too much force in the south which led to them getting routed in the north. If it had not been for (human controlled) Hungarian troops garrisoning and suppressing the western front, the liberation of France would likely be well under way. This is September of '41, six months after invading Russia.

 

[Meglok]

That would help a little, but the AI got through all those timing difficulties in good shape. The way it played was stupid but it got away with it. In fact the bigger blunder was the Allied landing in Germany within days of Poland being invaded. That cost them 40 divisions or so and greatly hastened the fall of France.

Changing the timing back might actually make things worse. If Weserubung had come sooner and Denmark had consequently been in the war before the fall of Poland many of those British units would have been able to fall back instead of being trapped against the neutral Danish frontier.

Is Poland really taking 70 days to fall? Right now, with Around Maginot firing 28 days after Danzig and Germany immediately DoWing BeNeLux Poland is still fighting and AI Germany is distracted by two expanding fronts. Iirc Poland usually lasted about 45 to 60 days in 1.3.2 when Germany just stared across the Maginot.

This might leave more units on "garrison" on the coast and discourage the AI from suicidal invasions. Although thinking about it I recall the allied AI planning those invasions as soon a it could without any apparent concern for AI German defenses. Obviously, the best outcome would be for PDS to fix whatever they broke when they coded the AI to "eat it's vegetables first".

 

[Dalwin]

I guess you are right. I forget the timing on those since I almost always fire those justifications manually instead of using the NF.

 

[wildbillhdmax01]

Came here just to post these.

Germany JUST declared war on Russia. This is what the front looks like. A handful of Divisions on the Russian front. There's more German Division in Africa then there is on the Eastern Front.... We are once again back to square one. Also German is suiciding it's Divisions in the English channel again.... This is my first game on 1.3.3., and... I just don't know..... I'm also seeing the suiciding of Navies.

 

[Gwydion5]

I don't really disagree with anything you're saying. But I would say my own point of departure as far as where my position begins to diverge from begins with that sentence above. And your more detailed explanation reveals how I have been consistently failing to properly explain my own position. The shuffle, the self-reinforcing nature, the fact that unit count can induce and/or increase this behavior...all true. But with AI Germany in particular the really game-breaking part, the persistent meltdown on the Eastern Front every single game, seems to begin with one simple issue: it is failing to properly determine what its needs actually are.

I agree with this as the fundamental problem. AI Germany not boosted, and AI Germany max boosted makes little difference to the Army/Front/BP Shuffle. The AI seems to lack a proper weighting mechanism to prevent it from make bad decisions. I've been human_ai observing Russia against a max boosted Germany and I have a couple screenshots I want to share.

War was declared late January, up to July it's been a rather back and forth fight. AI Russia is losing ground though slowly but surely. There are criticisms to be leveraged here on how and why. In fact I might take a crack at the January 1st 1940 save and see if I can't do better. But keeping on topic. AI Valuation judgement.

203 Divisions have been allocated to the southern western front, allowing AI Germany to just march on in Northern and Central Russia.

On the far west we have 60 divisions in a single province.

If there is no code in the game to stop the AI from doing things like this, then isn't the AI war just a matter of luck in terms of whichever AI does something like this first is likely to lose?

 

[Pro_Consul]

If there is no code in the game to stop the AI from doing things like this, then isn't the AI war just a matter of luck in terms of whichever AI does something like this first is likely to lose?

This might be so. The current fixation on Germany may be just the result that its greater strategic complexity, e.g. having more fronts to manage, gives it so many more opportunities to go off the rails.

 

[Gwydion5]

This might be so. The current fixation on Germany may be just the result that its greater strategic complexity, e.g. having more fronts to manage, gives it so many more opportunities to go off the rails.

Except this is Russia. As you said, the AI is failing to determine what its needs actually are. What need is there for 203 divisions on the southern front? What logic in the code concluded that it was a good idea to abandon Northern and Central Russia to overload the infrastructure in the south and cause (huge?) attrition to the divisions? I do not think it was coincidence that AI Germany aggressively captured territory in Northern and Central Russia. Even in the October 31st screenshot you can see northern Russia is lacking the necessary divisions to defend. But by December it seems like it has completely abandoned it, by placing 203 of 211 divisions in the south.

 

[Pro_Consul]

Except this is Russia.

I know. I was talking about the weather, so to speak. All the focus has been on AI Germany, as this thread's original premise bears out. I was just saying that the winds are blowing against Germany because they have many more opportunities to screw up and get fixated on phantom needs, but that this behavior is not limited to Germany by any means.