I did a really stupid thing

[kobayashi]

I decided to watch an all AI game of ZWK. So after activating the scenario flag, I set human player to none and pressed return a few thousand times over a few hours (including all the messages which require return). I suppose the idea was to see if the game was balanced and how well I could ochestrate proper AIs behavior without resorting to an evented script (think of it as an organic product ). I was quite pleasantly surprised.

It was something like a movie and I would sometimes stop for a while to see what the various powers were building and researching. The blitzkrieg went really well at the begining and the Axis bombers and tanks worked in unison to take over city after city. But by the time the other powers started getting AA guns, the He 111 and Ju 87 began to face heavy losses.

There was no battle of britain as Paris remained in French hands. This typically happens when the maginot line holds. In some earlier experiments, sometimes a few German bombers and tanks by luck attack sequentially and the line plus the rest of France falls. But not in this session.

At approximately the maximum expansion point (nov 1941), the Germans were still behaving as they were supposed to and started attacking Russia. Unfortunately, they were also fighting the Italians and that screwed up history. They simply did not have enough forces to get even close to Moscow. Another screw up was when the neutrals and minor allies started fighting each other.

Third Reich at its peak

 

[kobayashi]

Well the end came in December 1945 when the allies finally got to Berlin. The Russians made an earlier attempt but were stalled after getting to Konigsberg.

The Soviets had a good run in 1943 but their offensive ground to a halt when Jagdpanthers were built by the germans at an accelerating rate. As they did not have any good bombers, the only other gains they got for the rest of the game was in the south. However, they did do their part by getting to Ploesti and that really was the straw that broke the german's back (no more automatic veterans)

The Allies AI behaved really well and were flying about 20 bomber sorties per turn. This had a really devastating effect and there was little the germans could do until they started building the Me262 in the end. Each Me262 could take out two or three bombers but the sheer overwhelming number of bombers just could not be stopped. Over 200 4 engined bombers were lost.

The Germans finally researched the He277 bomber and built a few but too late. While researching King Tiger, Berlin Fell. At the time the Allies were building the manhattan project.

Allies enter Berlin

 

[kobayashi]

General Comments

German losses can be summarised at 230 armoured divisions and 284 Infantry divisions. 102 AA Batteries (95 lost)were built in response to the concentrated bombing effort by the Allies. Allies on the other hand lost 61 armoured divisions and 55 infantry.

As far as tank development, things went fine and every model was built. The germans favourite model was the Panther and its variant jagdPanther. Over 100 were built and they played a major part in holding off both the Russians and Allies until Ploesti fell. The Allies built very few of the earlier models but they wern't engaged in land battles till later so that is understandable.

Bombers were a heavy favourite of everyone and all models were built continuously throughout the war. Fighters were a different story. The Allies gave up on producing fighters after about a year even though the researched all the fighter techs. They also did not build the lightning although some Me110s were used in 1941. Me 110s were also the favourite of the Soviets. Almost a similar story for the Germans. When the FW190 was available all fighter production except for Me110 was stopped. But they did build the Me262 when it became available (Allies refused to build the Meteor even though they researched it). This problem with units of the 'air superiority' role is something I can't figure out.

One reason the germans lost was they wasted precious time researching some U-boat techs which the AI cannot use properly anyway. If I have a further version, I think I will make the U-boat techs much less desirable to the AI.

Originally posted by kobayashi

Losses were as follows (surviving units in brackets)

AXIS

Pz II 41 (2)
Pz III 17
Pz IV 43
Stug IV 26
Panther 52 (4)
Jagd Panther 47 (5)
Tiger 4 (6)
King Tiger 0

Me109 39
Me110 70 (1)
Me262 15 (1)
He111 255
He277 6 (6)
JU87 186 (7)

Infantry 255 (49)
Paratrooper 19 (4)
Grenadier 15
AA Battery 95 (7)
AT Battery 43 (11)
Pz IV Whirlwind 20 (3)

Cap Ship 16
U-boats 38


ALLIES

M3 Stuart 10 (12)
Matilda II 2
M4 Sherman 7
Churchill 3
Cromwell 6 (1)
M26 Pershing 24 (17)
JagdPanther 9 (11)

Spitfire 18
Me 110 12
Wellington 19
B17 4
Lancaster 93 (16)
He 277 35 (17)
B29 108
Typhoon 20 (13)

Capship 15 (14)
Sub Hunter 14 (18)
Carrier 6 (8)

Infantry 48 (142)
Paratrooper 7 (8)
AA Battery 3 (13)
AT Battery 43 (20)


Soviets

BT-7 10 (6)
KV-1 4
T34/76 33
T34/85 49 (29)
SU-100 58 (30)

I-16 37 (8)
He 111 108
He277 3 (31)
JU87 29 (25)
Sturmovik 150 (34)

Infantry 102 (134)
Paratrooper 0 (13)
AA Battery 14 (27)
AT Battery 21 (42)

Capship 12

 

[Marko]

It is not a stupid thing at all! I do it all the time
Use the barbarian view, then everything goes quicker and you don't need to hit Enter so many times.

 

[kobayashi]

But Marko, do you have ank input regarding the plane problem. Like do you use 'attack' or 'air superiority' for your fighters in TSFE. The AIs seem to be very willing to build any aircraft with the 'attack' role regardless of their stats.

I am thinking of changing everything to 'attack'. But fighters have very low attack factors in my scenario while bombers have very low defense factors. If the AI fighters start attacking ground units, they will be slaughtered.

p.s. the 'stupid' thing was not to use barbarian view. I wanted to watch the entire thing to see how the units were being used. I used the Axis 'view' since they were always attacking or being attacked. (of course this means I missed all the fringe battles but waiting 15 minutes per turn on a 'reveal all' was too much even for me.)

 

[SunTzu]

You need to make the fighters in ZWK more better at attacking ground units and sea units.
Like the british. The P-40 was a good ground attack fighter.
the Spitfire was more of an air to air fighter.....

 

[kobayashi]

Originally posted by SunTzu
You need to make the fighters in ZWK more better at attacking ground units and sea units.
Like the british. The P-40 was a good ground attack fighter.
the Spitfire was more of an air to air fighter.....

There are two types of 'fighters' in ZWK, air superiority fighters like Spitfire and ME 109 and ground attack aircraft like Stuka. Hawker Typhoon and Sturmovik become available at approximately the time the non-german powers realise the need for close air support. Air superiority fighters by definition are meant to fight in the air but will still have success at attacking soft ground targets like infantry (but not tanks and battleships).

There is a third type of fighter which is an escourt type fighter. They have slightly less powerful stats than the comtemporary air superiority fighter but increased range (4 vs 2) to match bombers. These would be the Me110 and P-38.

 

[Marko]

What is the plane problem? AI don't build fighters or don't use them correctly?

As much as I know, AI builds randomly all air domain units regardless their stats or roles. Same about sea units. (About land units I have found that things are even more complicated than is written in various scenario making guides).

Range 2 or more and air superiority role don't fit very well together for AI. To get effective anti-bomber fighters you have to make their range 1. For escort fighters attack role is OK, they end their turn usually on the same square as bombers, protecting them. Bad side is that they tend to behave exactly like bombers and attack cities etc. Good solution can be making their attack factor 0.

Hope that helps.

 

[kobayashi]

Marko,

AI builds fighters (move of 2) and use them properly (i.e. they don't attack tanks but go for bombers in the air).

When next generation fighter is available (which makes the first model obsolete) it doesn't build fighters anymore.

In the case of the axis when the third generation fighter is available (which makes the second model obsolete) it does build them. However, Allies don't.

I'm sure its a complicated thing related to other units (AA gun perhaps?) which become available cause each successive fighter has better stats but the same cost.

Also I wonder what effect the 'air superiority' role has since any unit that has the 'fighter' ability will go out and hunt for aircraft anyway. Maybe the 'air superiority' role makes units less likely to attack ground and sea units - but does it affact the AIs choice of build.

 

[Case]

I've read that giving units the air superiority role makes them more agressive (even for ground units without the 'can attack aircraft' flag), and its a good default role for agressive units (eg instead of 'attack')

 

[Marko]

I haven't met the fighter problem before. Perhaps because in TSFE units don't become obsolete but are replaced by newer ones.

I shall go home and look your scenario. Maybe I can find the solution.

AI changes too often Goto command for land units with "air superiority" role. So they are not good if you are using MOVEUNIT command. I have divided all offensive land units between attack and air superiority roles in order to force AI build more different units. But the most important units have "attack" role.

 

[kobayashi]

Originally posted by Marko

I shall go home and look your scenario. Maybe I can find the solution.

Thanks. I'm counting on you.

 

[Marko]

I looked these fighters in your scenario.

FW-190 fighter is in Warrior slot. It means that it can't be build by AI unless it has despotism form of government. It appears in AI build list but AI never choses it. Unit domain or stats are not important here.

P-38 fighters were built by computer in my game (I have MGE).
About P-51 and Meteor I found that if you set their role to defence then AI builds them.

Hope this helps.

 

[kobayashi]

Originally posted by Marko
I looked these fighters in your scenario.

FW-190 fighter is in Warrior slot. It means that it can't be build by AI unless it has despotism form of government. It appears in AI build list but AI never choses it. Unit domain or stats are not important here.

This is something new to me. I had thought the only government dependent unit slot was the fanatic. Does this mean that slot is completely useless? Maybe I'll swop the militia unit to that slot since it is a unit that exists at the beginning but is not built.

If you set the P-51 and Meteor role to defence, will they still go out and kill bombers in the air or just sit in the city and wait to be attacked? I think I am going to experiment with changing the stats incrementally and other things unitl they get built. Then I can figure out what is the trigger that stops them from being built. Maybe just for the sake of understanding how the AI works.

 

[kobayashi]

Well I didn't get to understand the nuances of the AI's choice of units but by setting the fighter's role to sea transport, I seem to have coaxed it into both building and using fighters. I also swapped the torpedo and FW190D slots so that should solve the warrior problem. Great!

I think I will release a version 2.1 after further testing.

 

[Marko]

Yes, warrior slot units are only good for human player, unless you set AI government to despotism.

I think that it is neccessary to make some thorough research about AI unit building choices. This would give invaluable help for us scenario creators.

I know that Xin Yu has made some research about it and I have also discovered few things. Maybe in future we can put all our knowledge together and write a scenario design tip?

 

[kobayashi]

Marko,

I discovered another quirk. The AI's behave differently in an AI vs AI situation as compared to an AI vs Human situation. For instance, from the beginning the British AI capital ships always bombard the Axis coast and land troops to capture Amsterdam and Rotterdam. They also invariably attack the independents.

But when a human is playing Axis, they are much less aggressive.

The strangest thing I saw was that the AI built the P-51 Mustang whereas in an AI vs AI game it would never do so.

I wonder what other factors are controlling the AI. Does the AI play differently against a human or is it just reacting to how the human plays. I tend to think the earlie is true.

 

[Marko]

As much as I know, AI diplomacy (war, treaties etc) is certainly different with human player than it is with other AI.

But I think that AI unit-building choices are not influenced by "human factor". At least my research haven't revealed any similar connection.
Differences can come because human player's behaviour (building-choices and tactics) is different and AI responses to this.

AI tactics against human is different in some cases (like using nuclear weapons) but mostly it seems to be controlled by same algorithms as against another AI.

I have read that naval bombardment algorithm (decision-making process to attack or not) takes account only relative strenghth of opposing units and presence of coastal fortress in the target city. (Coastal fortress factor was added in last patch before the release of FW. In original Civ2 version it didn't count it and ships made often suicidal attacks.)

So maybe the reason why the British AI is less aggressive against Axis human is that human protect his coastline better.

 

[kobayashi]

I found another exception to Mike Daumen's cardinal rule (as quote here)

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

It seems that the AI will build an inferior defend role unit even when a superior unit is available in the case where the superior unit is deemed too expensive for a city to build. Haven't figured out where the threshold is. In ZWK, the Axis AI will build both these land units:

Wehrmacht Div 12a 10d 1m (defend role -ignore walls)
AA Battery 16a 11d 1m (defend role - pike, aeigis & fighter)

 

[Marko]

Mike Daumen's rule is very interesting subject to research. It has many exceptions and it works sometimes quite unexpected way.

I think that I have figured all these nuances and their reasons out. After some days I'll release the artickle about AI building choices. It will cover everything what is connected with unit slots, domain, stats, costs, role and abilities and how it all influences AI building choices.

If you want you can help me very much. I started the second part of the research but alone I don't believe that I can finish it before autumn next year. It is about how global (diplomatic stance, military and economical situation etc) and geographical factors (city location etc) influence AI building choices in given city.

It is very practical for scenario builders because it is possible to force AI to build certain units in certain regions (by playing around with unit roles etc).