AI for MWIF - Poland

[Eichenblatt]

I think it is as soon as Russia is at war with Germany...right?[&:]

[Eichenblatt]

aaahhh.... I wasn't quick enough...

[composer99]

Yes, it is when USSR is at war with Germany. Patrice just put in the wrong power (no doubt because he was thinking of CW because he was just writing about how the CW gets some Polish units when USSR and Germany are at war provided the USSR has claimed Eastern Poland). [:D]

[Incy]

I suspect the approach used here is not a good one to arrive at a good polish defense.

The polish war is complex and involves many units, and a good defense will make detailed adjustments based on the exact position and strenght of german units.

I think a better approach to minor country setup would be something akin to:
-designate a number of predetermined setups (like we're doing here).
-simulate the attack impulse against each setup (using the major power AIO to attack each setup based on where units are located
-pick the setup that provides the worst outcome for the attacker (based on the probable outcome. Use the score the MP AOI calculates for picking the best attack for each possible defence)

This gets rid of the need to code a detailed 'recipy' algorithm for each minor country being attacked by each possible set of opponents. The AI would be based on simulation rather than algorithm.

This basic principle could be modified and refined, of course. For instance
-The AI could adjust setups after the fact by running the normal land move AI to check if the land move AI reccomends any land moves based on the situation in the chosen setup (could be applied recursively, until the land move AI doesnt want to move).
-The AI could modify the setup after the fact by trying to swap units with similar defense factors, to see if the new result was better

[Shannon V. OKeets]

ORIGINAL: Incy

I suspect the approach used here is not a good one to arrive at a good polish defense.

The polish war is complex and involves many units, and a good defense will make detailed adjustments based on the exact position and strenght of german units.

I think a better approach to minor country setup would be something akin to:
-designate a number of predetermined setups (like we're doing here).
-simulate the attack impulse against each setup (using the major power AIO to attack each setup based on where units are located
-pick the setup that provides the worst outcome for the attacker (based on the probable outcome. Use the score the MP AOI calculates for picking the best attack for each possible defence)

This gets rid of the need to code a detailed 'recipy' algorithm for each minor country being attacked by each possible set of opponents. The AI would be based on simulation rather than algorithm.

This basic principle could be modified and refined, of course. For instance
-The AI could adjust setups after the fact by running the normal land move AI to check if the land move AI reccomends any land moves based on the situation in the chosen setup (could be applied recursively, until the land move AI doesnt want to move).
-The AI could modify the setup after the fact by trying to swap units with similar defense factors, to see if the new result was better

Actually what you would want is to mini-max. That is, to examine the best possible moves for the opponent and then choose the defense that minimizes the maximum benefit for the opponent.

[wfzimmerman]

ORIGINAL: Incy

I suspect the approach used here is not a good one to arrive at a good polish defense.


I think a better approach to minor country setup would be something akin to:
-designate a number of predetermined setups (like we're doing here).
-simulate the attack impulse against each setup (using the major power AIO to attack each setup based on where units are located
-pick the setup that provides the worst outcome for the attacker (based on the probable outcome. Use the score the MP AOI calculates for picking the best attack for each possible defence)

This gets rid of the need to code a detailed 'recipy' algorithm for each minor country being attacked by each possible set of opponents. The AI would be based on simulation rather than algorithm.

Isn't this more or less what the AI is doing when it decides where to move and chooses attacks for a major power?

[Incy]

Yes and Yes (to both of you).
I think using a minimax simulation is easier and probably better, mostly because there are very, very many things that *could* play in significantly for minor setups. Trying to guess and then properly program and evaluate every possible factor that could affect a setup IMHO involves a huge amount of labour, and there is a high risk some key conditition will be left out (simply because it was to obvious, for instance). One example, what if the weather was Storm? Many of the AI recipies/setups we've made so far does not account for this, they assume fine weather.

Furthermore, the mechanism chosen seems to me to be separate form how the main AIO would have to function. The main AIO would need a way to determine every impulse where it wants its units, both longterm (where do I want my defence to be in several moves), and immediately (where do I want my defence *now*). Since the AI will need to be capable of this on a continous basis, it seems to me the same mechanism should be used for setup too. It would be bad to have two separate mechanisms (which would probably lead to the AI changing it's mind about where it preferred its units immediately after setup)

So what we need is a mechanism that can apply to any situation, at any time. But note that it could differ for different powers/countries (most easily by giving a single 'standard AI' a power sepcific map of valuable places/desired objectives/priorities/settings/whatever). I fear a rule based mechanism is not a good choice, because I'm sure it would requires a large amount of rules for a huge number of very specific palces/situations. But still there will always remain plases & situations that are not well defined as a rule set, and thus we're back at square one again, requiring a generic mechanism to handle everything that does not fit nicely into an identifiable ruleset (and again we get sudden changes in the AI as we move form one AI mechanism to a separate and different one).

So I'd rather see a AIO based on simulation/calculation, and that all we provide is heuristics. I.e. we tell the AI what it's goals should be and try to weigh those goals. This could be goals on many levels, and we would want standard ways of embedding goals into the game engine. Then the AIO simulates/calculates the best way to achive as many goals as possible as well as possible.

If we do this the recipies we're making must be different, they must *not* telle the AI how to achieve something, they must only tell the AIO what it wants done.

Example:
-rather than telling the portugese AIO where it should place its units so that key hexes won't be easily invaded, we should let the AIO know what hexes are important to protect, and then let the AI figure out by itself where to place units to achieve that. Note that we'd have to do this in some sort of dynamic way. Coastal hexes, for instance, are much more valuable if no nearby hexes are enemy held, removing the need for the enemy to invade.
-rather than instructing the AIO on how to implement marauder setups, we should tell the AIO where key enemy hexes are and how much they're worth, and then the AI can calculate the odds of capturing each objective by simulating moves and/or through specialised sub-AI's suited to evaluate this problem.

[Shannon V. OKeets]

ORIGINAL: Incy

Yes and Yes (to both of you).
I think using a minimax simulation is easier and probably better, mostly because there are very, very many things that *could* play in significantly for minor setups. Trying to guess and then properly program and evaluate every possible factor that could affect a setup IMHO involves a huge amount of labour, and there is a high risk some key conditition will be left out (simply because it was to obvious, for instance). One example, what if the weather was Storm? Many of the AI recipies/setups we've made so far does not account for this, they assume fine weather.

Furthermore, the mechanism chosen seems to me to be separate form how the main AIO would have to function. The main AIO would need a way to determine every impulse where it wants its units, both longterm (where do I want my defence to be in several moves), and immediately (where do I want my defence *now*). Since the AI will need to be capable of this on a continous basis, it seems to me the same mechanism should be used for setup too. It would be bad to have two separate mechanisms (which would probably lead to the AI changing it's mind about where it preferred its units immediately after setup)

So what we need is a mechanism that can apply to any situation, at any time. But note that it could differ for different powers/countries (most easily by giving a single 'standard AI' a power sepcific map of valuable places/desired objectives/priorities/settings/whatever). I fear a rule based mechanism is not a good choice, because I'm sure it would requires a large amount of rules for a huge number of very specific palces/situations. But still there will always remain plases & situations that are not well defined as a rule set, and thus we're back at square one again, requiring a generic mechanism to handle everything that does not fit nicely into an identifiable ruleset (and again we get sudden changes in the AI as we move form one AI mechanism to a separate and different one).

So I'd rather see a AIO based on simulation/calculation, and that all we provide is heuristics. I.e. we tell the AI what it's goals should be and try to weigh those goals. This could be goals on many levels, and we would want standard ways of embedding goals into the game engine. Then the AIO simulates/calculates the best way to achive as many goals as possible as well as possible.

If we do this the recipies we're making must be different, they must *not* telle the AI how to achieve something, they must only tell the AIO what it wants done.

Example:
-rather than telling the portugese AIO where it should place its units so that key hexes won't be easily invaded, we should let the AIO know what hexes are important to protect, and then let the AI figure out by itself where to place units to achieve that. Note that we'd have to do this in some sort of dynamic way. Coastal hexes, for instance, are much more valuable if no nearby hexes are enemy held, removing the need for the enemy to invade.
-rather than instructing the AIO on how to implement marauder setups, we should tell the AIO where key enemy hexes are and how much they're worth, and then the AI can calculate the odds of capturing each objective by simulating moves and/or through specialised sub-AI's suited to evaluate this problem.

While the AIO could do what you describe, more or less, for setting up the minors, it is a heavy handed approach. I don't blieve it is necessary to run through a complicated set of code to decide where to place one or two units.

For example, the AIO will usually be making decisions about movement and the positions of units during play (not during setup of a minor) based on an overall evaluation of the worldwide situation, the balance of power in each of the branches of the armed forces for each side within each theater of operations. Besides this assessment, it will also be using strategic plans about where to attack/defend, both immediately and within the next year.

To go through all this analysis to place a couple of units on the map is excessive. Furthermore, I do not want to integrate additional logic into the master routine to handle the task of placing units on the map. In fact, bringing in reinforcements will be handled separately as well.

I guess my point here is that a single general purpose routine is not always the best solution. As the problem space expands to include different/unique situations, such as setting up minor countries and bringing in reinforcements, the logic grows more complex. It is better to cut the problem into separate pieces whenever possible. If the need arises to perform similar calculations in different places in the logic, then a 'subroutine' can be created to process the common logic.

To strain an analogy, there is no need for a large commercial airplane when a piper cub can do the job just as well.

---
Weather for movement is handled by the logic within a LAIO routine called 'Reach'. This tests whether a unit can 'reach' a given hex, and allows for such fine points as maybe being disorganized upon arrival, and having to skirt neutral hexes. This is easy to do since there are routines within the MWIF code that already perform these calculations. LAIO simply converts the Reach function into a call to the comparable existing function in MWIF code (e.g., for land, air, and naval units).

[peskpesk]

* Ground strike defence B

[peskpesk]

The plan behind Ground strike defence. In the B variant might even the 3-4 reinforce Warszawa, should it have survived the first attack. Remember that setup is only possible if using with Motorized movement rates and not with railway movement

[Incy]

Just a comment on the GS defence setup, id requires the CW to take a combined or land in it's first impulse.
Very often the CW wants to take a naval on its first impulse, particularily if france first is happening.
So this setup, while very good, must be seen in the context of the global situation for the CW, because it consumes action limits.

[peskpesk]

ORIGINAL: Incy

Just a comment on the GS defence setup, id requires the CW to take a combined or land in it's first impulse.
Very often the CW wants to take a naval on its first impulse, particularily if france first is happening.
So this setup, while very good, must be seen in the context of the global situation for the CW, because it consumes action limits.

Excellent! I will add the demand for CW to do combined or land in it's first impulse to use the setup.

[Taxman66]

One of the things we sometimes do (did) was to fly the polish airforce into the Baltic States on the CWs first impulse if GE set up enough FTRs to make it worthless to do anything else.  This allows the CW to get 2 pilots without having to wait for GE and RU to go to war.  The Polish planes go away, but the CW really doesn't want them anyway.

[Froonp]

ORIGINAL: Taxman66

One of the things we sometimes do (did) was to fly the polish airforce into the Baltic States on the CWs first impulse if GE set up enough FTRs to make it worthless to do anything else.  This allows the CW to get 2 pilots without having to wait for GE and RU to go to war.  The Polish planes go away, but the CW really doesn't want them anyway.

I do this all the time, not just sometimes.
Becoming CW pilots will give them the opportunity of flyin lots of missions, as the CW has lots of unpiloted planes, rather than fly once in Poland for nothing and get captered.

[Taxman66]

The key is that GE has to set up 3 or 4 FTRs on the Polish Border, if he sets up only 1 or 2 then it might be worth using the Polish airforce.  Particularly if they can fly from the target hex back to an Eastern Polish hex or (unlikely) to the Baltic States.

[Orm]

Nowdays I almost always fight with the Polish aircraft.

The reasons that I do this is:
-You force the German planes to commit on the Polish front.
-Makes Germany spend oil to reorganize his planes.
-A chance to shoot down German aircraft and to alter land combat odds.
-A decent chance for the Polish pilots to survive to join the CW. (If the aircrafts is shoot down)

Remember that a pilot in a aircraft shoot down over friendly territory is more likely to survive than to be killed.


-Orm

[Taxman66]

Yeah, but when GE is rolling on the big plus tables there's a pretty good chance the pilot is going to die...  You also have to remember that if the Polish plane(s) is only aborted the pilot is going to die on the ground (remember you can't fly on GE's first impulse) or possibly be held in Siberia (if RU hasn't claimed E. Poland yet.  In our games this happens right away as you never know when a S/O will end.  Even still a pilot held in Siberia isn't doing the CW any good.

[Orm]

ORIGINAL: Taxman66

Yeah, but when GE is rolling on the big plus tables there's a pretty good chance the pilot is going to die...  You also have to remember that if the Polish plane(s) is only aborted the pilot is going to die on the ground (remember you can't fly on GE's first impulse) or possibly be held in Siberia (if RU hasn't claimed E. Poland yet.  In our games this happens right away as you never know when a S/O will end.  Even still a pilot held in Siberia isn't doing the CW any good.

USSR is always guaranteed at least 2 impulses in sep/oct 39 so Russia do not need to claim eastern Poland on the first impulse.

If a plane gets shot down over friendly terrotory the odds for the pilot to survive is always alot better than for the pilot to die. Regardless of what table that has been used.

The only thing that worries me with this strategy is the abort. And then we played that a pilot interned by the russians was removed from the game.

Since the rule is: (My bold markings)
19.5.1 Eastern Poland
Any Allied (except Soviet) units there are destroyed.They are removed from the game (internment) until Germany and the USSR are at war, at which point they may be added to the Commonwealth force pool if the Commonwealth player so desires.

-Orm

[Taxman66]

A 17-20 (+2 to +5 GE advantage) will kill the Polish Pilot.  Additionally 7, 8, 13 & 14 (or 8, 9, 13 & 14) will get the pilot killing abort; and a dash or even a Clear through by the GE could result in the GE ending the fight providing an 'in effect' abort result to the Polish. 
 
If GE commits only 1 or 2 FTRs to the Polish border it 'may' be worth flying the Polish air force.  If does have them, then I prefer the guarantee to get the 2 pilots rather than the small chance of doing harm to the GE forces.

[Froonp]

ORIGINAL: Orm
The only thing that worries me with this strategy is the abort. And then we played that a pilot interned by the russians was removed from the game.

Since the rule is: (My bold markings)
19.5.1 Eastern Poland
Any Allied (except Soviet) units there are destroyed.They are removed from the game (internment) until Germany and the USSR are at war, at which point they may be added to the Commonwealth force pool if the Commonwealth player so desires.

-Orm

No, the interned pilot is not destroyed, if the plane aborts to Eastern Poland and Eastern Poland is controlled by the USSR, the Polish Pilot is interned by the USSR, and may be added to the CW pilots when the USSR comes to war with Germany (as per the rule you quoted).

This said, aborting to a neutral minor country is better, but is there a neutral minor country in range from the place where the Polish Air Force will engage the Germans ? I doubt it, so I prefer rebasing them immediately to a neutral Minor Country for the immediate gain of 2 pilots. The CW desperatly needs more pilots to pilot its 4 unpiloted bombers.