Improving AI Germany vs AI USSR in 108

[Denniss]

With pang's save and the old debug .exe I get some error messages but ignoring is possible
Assertion failed: pUnit->IsDeleted() == false file:combat.cpp line:2188

I'll do some tests with the latest standard .exe. It's somewhat odd to see Cuba with 11k money in the negative though.
No crashes until I stopped in October 1941, autosaves set to monthly.

 

[Pang Bingxun]

Another important thing to note is that reloading a save from mid july makes all those Arm suddenly move from the french beaches to the soviet theatre.

It might not be very surprising that germany performs barbarossa much better when it uses all fast units for it. But saving the game at the start of the war and reloading the save helps germany a lot. Reloading does increase the soviet losses in the first 30 days from about 20 divisions to 60 divisions. This reloding issue seems to be a serios bug. Not reloading makes the ai not do the necassary reevaluation of needs. And reloading makes the ai stop moving units, doesn't it?

I reloaded the save i attached above from june the first. After 60 days germany made impressive advances. About 110-120 soviet divisions were destroyed by being overrun:
Germany: 52 Inf, 1 Cav, 36 Mot, 12 l.Arm, 13 Arm, 4 Par, 15 Mtn, 5 Gar, 4 HQ, 142 Divs. 438/283 IC. 17901.7 rares at -60.19 per day
USSR: 132 Inf, 4 Cav, 33 Mot, 5 l.Arm, 14 Arm, 6 Mtn, 2 HQ, 196 Divs. 264/188 IC
So on august the first germany controlls provinces neighboring to moscow and german troops are advancing into Stalingrad where the defenders are already beaten.
On august the 12th the last gpw event fires. On august the 18th moscow falls. At the end of august i have a CTD.
Anyway this is a huge dependence of the german ai on the human saving behavior.

 

[unmerged(86927)]

I restarted with the save from may the 21st but the new exe.
Barbarossa starts again on may the 22nd. I thought i had reduced the chance back to 20% but as i take a look on the event i see it is still 50%. I guess i change it back to 20%.
In early july i get a CTD altough i did make no attempt to save.

I restart again. Barbarossa starts on june the 1st. A few days later it crashes.

I restart again. In early june it crashes before barbarossa starts.

I restart again. now barbarossa starts on june the first. The chance are 10:90, still barbarossa often starts on une the first. I save the game.
At the end of july i get a CTD. I hope this save is of some help.

The save game you attached was useful. I have uploaded a new exe. Please report the situation of crashes with it.

 

[unmerged(86927)]

This reloding issue seems to be a serios bug. Not reloading makes the ai not do the necassary reevaluation of needs. And reloading makes the ai stop moving units, doesn't it?

While not having the time to study the mechanics behind, I have noticed this on Spanish civil war too. It usually gets stuck for years until you reload and it then it progress.

 

[Pang Bingxun]

I reloaded the save i attached above from june the first. After 60 days germany made impressive advances. About 110-120 soviet divisions were destroyed by being overrun:
Germany: 52 Inf, 1 Cav, 36 Mot, 12 l.Arm, 13 Arm, 4 Par, 15 Mtn, 5 Gar, 4 HQ, 142 Divs. 438/283 IC. 17901.7 rares at -60.19 per day
USSR: 132 Inf, 4 Cav, 33 Mot, 5 l.Arm, 14 Arm, 6 Mtn, 2 HQ, 196 Divs. 264/188 IC
So on august the first germany controlls provinces neighboring to moscow and german troops are advancing into Stalingrad where the defenders are already beaten.
On august the 12th the last gpw event fires. On august the 18th moscow falls. At the end of august i have a CTD.

I resume the save from august the first.
On august the second the last gpw event fires. On august the 13th moscow falls.
2 days later the germans take the new capital Naberezhnye Chelny and a lot of oil and rares.
On september the first i count 146 german divisions vs. 179 soviet ones. I save the game and get no crash.
oil = 60225.9883 and rare_materials = 22784.8125 i read in this save for germany. The daily loss of rares is down to 39.59.
On october the first i count 147 german divisions vs. 173 soviet ones.
On october the 24 Leningrad falls. As Naberezhnye Chelny has been retaken by the soviets this does not give germany any resources.
On november the first i count 149 german divisions vs 148 soviet ones. Norway and italy are under attack by the UK. I save.
In november germany loses its initial advantages. In terms of IC germany is superior. 443/286 IC vs. 187/133 IC is quite clear.
On january the first japan starts the war with the USA. I save and reload.
On january the 15th i count 159 german divisions vs 156 soviet ones. This does not include that many german arm division have lost about 50% strengths and cannot be repaired fast due to a low ESE.
On january the 18th germany takes Naberezhnye Chelny and some good stockpiles.
On february the first italy is down to 9 divisions. South italy and greece are under control of UK forces. Baku is controlled by germany.
I count 162 german divisions vs. 152 soviet ones. i save.
In the mid of march Archangelsk and Kem are taken. 26 soviet divisions are now cut of from the main forces.
On april the first i count 175 german division vs. 146 soviet ones. The airforces: germany= 76 scadrons, USSR=55, USA=55, JAP=27.
On may th 16th the Kola pocket is finished. Due to new ai events it is now controlled by finnland that now controls everything it would get from the bitter peace event.
On june the first in 1942 i count 178 german divisions vs. 81 soviet ones. I save. No crash within 10 month.
In august the second germany has less than 200 manpower. Joseph Goebbels and Ernst Busse will take power on august the 17th.
On october the first germany has 71 inf, 1 Cav, 47 Mot, 3 Mech, 13 l.Arm, 19 Arm, 4 Par, 15 Mtn, 22 Gar, 7 HQ, 202 Divs; 460/288 IC.
On november the 26th germany is short on rares. the IC now fluctuates between 130 and 460. Really strange.
On january the first in 1943 germany has 208 divisions, the soviets have 80. A standoff is reached. I save.
I reload as germany and see no ic on reinforcements. Production is prefered instead.
The AI is poor at handling fluctuating ic. The AI Priorities seem to be opposite of what is needed. Production must have the lowest priority:

1. Infrastructure repairs. This increases IC and must be the top priority.
2. Supply production. As lowness on supplies will cause additional need for reinforcements. Supplies are needed in a proper amount.
3. Reinforcements of troops. This needs a high priority as reinforcing immediatly will save a lot of ic.
4. Consumer goods. Money is needed in a proper amount.
5. Upgrading. This needs a medium priority.
6. Production of new units. This is the most expensive and the least efficient way to increase the fighting capabilities, so it needs a low priority. An exception could be additional infrastructure in areas of too low infra, but the ai will probable not manage this.

I attach a save. Maybe it helps fixing the ai in such crucial situations.

BTW: With the new exe i had no crashes yet.

 

[PB-DK]

all axis nations have troubles with resources, they do not build up any real stockpiles before the war nor do they build any converters that i have noticed after they run low

 

[Pang Bingxun]

Here is a new try. Barbarossa starts on june the 14th.
Germany: 57 Inf, 1 Kav, 35 Mot, 12 l.Arm, 14 Arm, 4 Par, 14 Geb, 6 Gar, 4 HQ, 13 MRF, 21 Int, 10 CAS, 15 Tac, 5 Nav, 3 Tra, 436/281 IC.
USSR: 221 Inf, 5 Cav, 40 Mot, 9 l.Arm, 23 Arm, 11 Geb, 8 HQ, 8 MRF, 19 Int, 13 CAS, 15 Tac, 348/249 IC.
On july the 14th the number of soviet divisions has dropped from 317 to 248. the number of german divisions has dropped from 147 to 146.
On july the 27th the forth gpw event fires. This reduces the dissent. USSR has 201 divs left and from now on the rules are playing much less in german favour.
On august the 11th the fifth gpw event fires. So the soviet do now only suffer -10 max org.
On august the 14th i count 149 german divisions vs. 188 soviet ones. The german IC has dropped to 291/280 IC, the rares are rare again.
Is there a way to increase rare stockpiling? Is there a way to increase infra production? If i simply increase max factor, than more factories than supplyable are build.
On september the 8th moscow falls. On september the 14th i count 150 german divisions vs. 175 soviet ones. On september the 30th the last gpw event fires.
On october the 14th i count 151 german divisions vs. 158 soviet ones. 63 of them are in the big Leningrad-Murmansk pocket.
Is there are way to make the soviets weaken the finnish front in favour of the main front? The ratio seems quite irrational.
On november the 13th i count 151 german divisions vs. 153 soviet ones.

The soviet airforce might be increased at the expense of the army. The german total army strenghs might be increased at the expense of the expensive units. More Inf and more GAR are needed. Also a different building skrip if germany controlls much soviet territory might help with more Gar when needed.

 

[unmerged(86927)]

Why do you intend to increase the strength of the German army vs Soviet AI since now I see Moscow falling? The purpose of changes was to make German AI advance more than a couple of provinces in depth, not steamroll the entire European Russia, taking all major cities.

 

[NiL_FisK|Urd]

Now its time to increase the soviet defense, then again the german offense, and after a few cycles tha ai could be a challenge for human players ^^

 

[Pang Bingxun]

Why do you intend to increase the strength of the German army vs Soviet AI since now I see Moscow falling?

The soviet defence is artificially weakened by dissent. I could remove that very easily to any degree i like, so i donnot worry about the soviets. But the idea to increase german numbers is not necassarily going to incease their performance as increasing numbers means decreasing the number of quality troops.

 

[Denniss]

Please reduce the still enormous amount of german mot inf divs - let them build mor GAR divs to free up other divs from garrison duty.
Does GER really needs to have that many MRF/INT/TAC ?
If GER builds too many IC change "ic_end_year" to an earlier date and/or remove the least effective IC provs from the list (those with low infra). Innsbruck Graz, Luxembourg/Metz/Strasbourg/Mulhouse
If GER runs out of rares too often please increase their value in the trade ai - in CORE we use 6 for rare_materials in GER ai (default value is set to 4)

 

[NiL_FisK|Urd]

If GER runs out of rares too often please increase their value in the trade ai - in CORE we use 6 for rare_materials in GER ai (default value is set to 4)

Does this really help? I thought this only affects the price the ai is willing to pay for it.

 

[unmerged(86927)]

The soviet defence is artificially weakened by dissent. I could remove that very easily to any degree i like, so i donnot worry about the soviets. But the idea to increase german numbers is not necassarily going to incease their performance as increasing numbers means decreasing the number of quality troops.

Maybe now you can restore tank destroyer tech unavailable for Germany at the start of the game, as it was before

 

[Pang Bingxun]

Please reduce the still enormous amount of german mot inf divs - let them build mor GAR divs to free up other divs from garrison duty.
Does GER really needs to have that many MRF/INT/TAC ?

I intend to change that. The too high number of quality troops is is side effect of changing to relative buildung scheme but still using the old numbers. This also tends to make other UK and others to build too much armour. Maybe armour allocation should be reduced in early years and increased after the start of the year?

If GER builds too many IC change "ic_end_year" to an earlier date and/or remove the least effective IC provs from the list (those with low infra). Innsbruck Graz, Luxembourg/Metz/Strasbourg/Mulhouse

Reducing the number of prefered ic provs does not help. If there are no prefered provs left than the ai will simply build ic in non prefered provs with low resource production. It would be nice to make the ai use the concentration bonus. This could be achieved by making the ai build ic in prefered provs only unless told otherwise.
But right now the usage of a lower ic_end_year seems the best alternative.

If GER runs out of rares too often please increase their value in the trade ai - in CORE we use 6 for rare_materials in GER ai (default value is set to 4)

Does this help? It did not help for supplies in case of CHI.

Maybe now you can restore tank destroyer tech unavailable for Germany at the start of the game, as it was before

If you are refering to sp_art than intend to do so. But maybe the start year could be reduced and the requirement be lowered to 1936 Art instead of 1938 Art. My idea is to use relative buildung scheme to build only airforce and ships before the sp_art is available and as soon as sp-Art is researched attachment preferences, building scheme and max_factor are changed to make germany build a strong big army.

 

[Denniss]

A higher setting for rare in the trade ai should get the ai to increase the number of rare trades and probably also increase how much an AI would pay for it. the ai doc file does not really have a lot of information, maybe gunman could have a look into the real effects of these value changes.
I don't know how the trade situation is for GER but if they are trading away oil/rares the ai doc file offers a trade block setting.

 

[Pang Bingxun]

germany does not export rares and exports 0.3 oil only to finland for some rares. Altough i did check for the beginning of 1942 only blocking exports seems unneeded. The problem is mostly that german stockpiles low amounts of rares while hungary has more than needed. The stockpiles of oil on the other hand seem almost too big.

 

[bestmajor]

i didnt follow this whole thread, but in case germany looses vs SU, maybe let SU research naval techs as well? they were building some ships before 41 iirc and this will slow them down with inf or doctrine techs?

 

[Pang Bingxun]

Here is a new try with beta6. barbarossa starts on june the 3rd.
Germany has 64 Inf, 1 Cav, 38 Mot, 3 l.Arm, 16 Arm, 4 Par, 14 Mtn, 63 Gar, 5 HQ, 208 Landdivs, 12 MRF, 19 Int, 11 CAS, 13 TAC, 5 Nav, 3 Tra, 63 Airdivs; 483/311 IC. So far so good.
Soviet union has 192 Inf, 5 Cav, 32 Mot, 4 l.Arm, 31 Arm, 11 Mtn, 9 HQ, 284 Landdivs, 15 MRF, 13 Int, 15 CAS, 15 Tac, 3 Tra, 61 Airdiv; 355/245 IC. Too much HQs and Arm.
From murmansk till Viipuri the USSR has 5+3+3+3+3+3 Divs. 20 Divs at the finland front is imo a bit too much, but that is better than in Earlier versions. Finland has 19 divs.
On june the 25th the third gpw event fires. I count 208 vs. 263 divisions.
On july the 9th the fourth gpw event fires. I count 209 vs. 230 divisions.
On july the 23th the fifth gpw event fires. I count 209 german divisions vs. 204 soviet ones.
On august the first Leningrad is fallen. I count 30 soviet division at the finland front.
On september the first Stalingrad is fallen. I count 213 vs. 190 divisions. The soviet split finnland into north and south.
On september the 12th germany has less than 300 manower, Ernst Busse will take power on september the 27th.
On september the 26th the last gpw event fires. I count 216 german divisions vs. 187 soviet ones.
I think it is time to remove some of the soviet restrictions because moscow has fallen again.

 

[Pang Bingxun]

I resume my save from june the third.
The third gpw event fires on june the 25th. I count 257 vs. 208 divisions.
The fourth gpw event fires on july the second.
The fifths gpw event fires on august the 14th. i count 211 vs. 210 divisions.
On september the 15 Stalingrad falls, but Geyr von Schweppenburg is pocketed there. A few days later Leningrad falls.
On October the 10th the lst gpw event fires. I count 214 vs. 181 divisions.
On november the 6th germany controlls 4 provinces adjacent to moscow. So 25 divisions could attack the 3 in moscow, but they donnot.
On january the first i count 220 german divisions vs. 183 soviet ones. Things begin to solve off quite well.
Still the soviet ratio of finland front to german front seem a bit too high. The ai should not use 30 division against 19 finnish divisions.

 

[Denniss]

Still way too many german mot divs and not enough inf thus GER is advancing way too fast.. Regarding FIN please expiriment with a FIN country priority set to 10 or 15 in your save, sov_fin_demands.ai and sov_germany.ai