Strategy 101 resources wanted

[ralphtricky]

ORIGINAL: sPzAbt653

He should not be overstacking, I'll look at that and the flanking. Part of that is that he can be much less cautious depending on the formation stance.

I just realised that changes may have been made since 3.4 or early 3.5, so I may be using an outdated Elmer. Sorry about that.

I don't think it has changed that much. I guess the question is what do you mean by overstack? He DOES overstack when prosecuting an attack on the most aggressive level. Other than that he does stack, but tries not to exceed the limit where the overstacking indicator appears.

[ralphtricky]

ORIGINAL: golden delicious

ORIGINAL: ralphtrick

Here's a good talk about the difference between 'good' and 'fun' AI. Eventually, I also want to add 'personality' to the AIs, some might be more aggressive, some might be more passive. Usage of armor for breakthroughs, different levels of reserves might be kept, etc.

http://www.youtube.com/watch?v=IJcuQQ1eWWI&feature=sub

IF year = 1900 AND playerwinning = "Yes" THEN DOWHuman()

Wow.

If you're losing and close to the end of the scenario, forget history and see if you can take anything.

Code: Select all

 if (Events.winner() != currentPlayer) {
 	if (OPCalendar.calendar.currentTurn > (OPCalendar.calendar.finalTurn - 3)) {
 		generalOrdersBias = BANZAI;
 		generalEmphasis = ALL_COSTS;
 		poAttackFlavor = ATTACKING_STATUS;
 		poLog(logPOGeneral, " PO general orders: Banzai.");
 		poLog(logPOGeneral, " PO general emphasis: Ignore losses.");
 	}
 }
 

[ralphtricky]

ORIGINAL: governato
Ralph,

There is an hour long lecture from Dave O'Connor, (Arjuna), the developer of Command Ops here:

http://www.wargamer.com/hosted/dropzone ... index.html

You are probably familiar with the description of the game, but the audio file on the AI algorithms is quite interesting and the short articles are hopefully useful to improve the AI a bit.

Thanks, I'll take a look.

[ralphtricky]

The current code tries to help the AI make good decisions, but it may make it a little harder for people to program it. It does this using the total AP force on the map, then it does something else for each of the formations to modify it based on all VISIBLE units within about 9 hexes of the center of the formation. It uses this final multiplier to determine how the AI reacts.

This initial check is the one that I wonder if I shouldn't just eliminate (except for the final push[:D].) I understand why it's there, and how it can help, but it seems like it makes it harder for designers. The other option I might pursue is to treat the designer value as an override instead of a modifier if set to anything other than 'stand'.

[sPzAbt653]

ORIGINAL: ralphtrick

ORIGINAL: sPzAbt653

He should not be overstacking, I'll look at that and the flanking. Part of that is that he can be much less cautious depending on the formation stance.

I just realised that changes may have been made since 3.4 or early 3.5, so I may be using an outdated Elmer. Sorry about that.

I don't think it has changed that much. I guess the question is what do you mean by overstack? He DOES overstack when prosecuting an attack on the most aggressive level. Other than that he does stack, but tries not to exceed the limit where the overstacking indicator appears.

This is an example using 3.4. As the German (grey) units are on the defense, they are in a good terrain position (the mountians to the southwest have badlands, so they are good flank protection). But, they are too dense, on minimal losses and not dug in. It seems to be a good defensive position, but with a horrible defensive unit arrangement. Can Elmer make a determination like 'if on defense, and stacked in the red, go for some depth' ?

[sPzAbt653]

(OPCalendar.calendar.finalTurn - 3))

I know there is good reason for this, but it usually means that Elmer plays better on the final three turns. I've tried setting him to Attack and Berzerk but he still doesn't go for it unitl the last three turns. Is there a way to trick him into thinking its that late earlier in the scenario ?

[sPzAbt653]

ORIGINAL: ralphtrick

The current code tries to help the AI make good decisions, but it may make it a little harder for people to program it. It does this using the total AP force on the map, then it does something else for each of the formations to modify it based on all VISIBLE units within about 9 hexes of the center of the formation. It uses this final multiplier to determine how the AI reacts.

This initial check is the one that I wonder if I shouldn't just eliminate (except for the final push[:D].) I understand why it's there, and how it can help, but it seems like it makes it harder for designers. The other option I might pursue is to treat the designer value as an override instead of a modifier if set to anything other than 'stand'.

I don't understand enough of this to make a useful comment, but the 'VISIBLE units within 9 hexes' thing is something I think Soren Johnson discussed. He said that a human player will generally know the probability of a unit being in a location even if its not visible, and in may cases it is therefore 'ok' to let the PO see units that have been seen previously. I think that's what he said.

[Grognard]

It seems to be a good defensive position, but with a horrible defensive unit arrangement. Can Elmer make a determination like 'if on defense, and stacked in the red, go for some depth' ?

This example illustrates some of Elmer's common tendancies:

Some points might be illustrated (or argued) in this example. (Let's assume its Elmer vs Elmer)

1. Not dug in - bad Elmer/not drilled Elmer. A unit not planning to move the rest of the turn should/must always dig in. No arguements. Even with clear superiority.
1a. Min losses - Not sure why - Elmer is weak at risk/reward
2. The Panzer Rgt stack doesn't even need to be there as the other Elmer would never push a unit into the subsequent gap. Never...
3. The red stack might be (IMO "is") there to keep the odds <2:1 in which case Elmer knows Elmer won't attack.
4. I'm betting there's Arty under the Army HQ - otherwise Elmer wouldn't have assigned his SP unit to Tac Rsv. (assuming Arty is range 5 )
5. Regardless of recon level (even 0) Elmer knows that Army Arty is there. (My tin hat paranoia)
6. Corollary of 4 & 5. - If no Arty under Army HQ I wonder if the SP unit would be on Tac Rsv. - not necessarily...

Like I said - open to argument - but it generally has held true in my PO games. Operative word is "generally"

And actually - the above example is a little better than his usual behavior. I probably would not assault that position w/o a few more assets. Or a boatload of Arty under that HQ

[sPzAbt653]

A unit not planning to move the rest of the turn should/must always dig in.

I'd take it a step further and say that any unit that will stay in place should dig in, but I don't know if Elmer ever determines that a unit with movement allowance left will not move. I would guess that as the situation may change due to friendly combat, he is always leaving his units available to move. As the human I always dig in if possible (why stand around if you can take cover?).

Elmer would never push a unit into the subsequent gap.

Very true, discouraging Elmer from doing the pp dance has made him not want to move thru gaps covered by zoc's (even untis with 99% recon).

assigned his SP unit to Tac Rsv. (assuming Arty is range 5 )

Range is 4, but maybe he couldn't move closer that turn.

the above example is a little better than his usual behavior. I probably would not assault that position w/o a few more assets.

Yes, but we are offering suggestions to Ralph that might help improve Elmer. One question is, as a player, could you set up a better defense, and if so could this also be made obvious to Elmer. Another question would be is the 'overstacking' necessary. I would think defense in depth would be better. What would other players do if facing a human opponent ?

[Grognard]

Another question would be is the 'overstacking' necessary. I would think defense in depth would be better. What would other players do if facing a human opponent ?

Exactly; this illustrates the difference 'tween Elmer & us. Against Elmer I will overstack to get the odds under 2:1 because I know Elmer won't attack. Against a Hu-man - never, never, never - defense in depth is more tactically sound. Reversing the roles, I would certainly do a frontal assault against a red stack at less than 2:1 just to produce the excess casualties and I would expect any Hu-man opponent to do the same. Whereas if I see your defense in depth (even with > 2:1 odds for the initial attack), a possible counterattack would certainly give me pause in the example above.

Would it be possible to program Elmer with a chance variability in his attack routine? i.e. instead of Elmer just crunching numbers - if certain conditions are met (red stacks, objective hex, key terrain hex) give him a binary yes to attack where he normally wouldn't.

To what extent does Elmer currently recognize key terrain? Chokepoints, flank anchors, etc.

[ralphtricky]

This is an example using 3.4. As the German (grey) units are on the defense, they are in a good terrain position (the mountians to the southwest have badlands, so they are good flank protection). But, they are too dense, on minimal losses and not dug in. It seems to be a good defensive position, but with a horrible defensive unit arrangement. Can Elmer make a determination like 'if on defense, and stacked in the red, go for some depth' ?

Probably, one issue is that the only defense that Elmer knows right now is a a rolling defense, not a static one. I'll take a look at what he does on defense so see if I can stiffen it.

[ralphtricky]

ORIGINAL: sPzAbt653

(OPCalendar.calendar.finalTurn - 3))

I know there is good reason for this, but it usually means that Elmer plays better on the final three turns. I've tried setting him to Attack and Berzerk but he still doesn't go for it unitl the last three turns. Is there a way to trick him into thinking its that late earlier in the scenario ?

I am probably going to rework that for 4.0, right now, there is also a reliance on the overall odds which make it more difficult to determine that. I know there there is a global value for each side that controls Elmer's aggressiveness which should normally do what you want, although I don't remember what it's called.

[ralphtricky]

ORIGINAL: sPzAbt653

ORIGINAL: ralphtrick

The current code tries to help the AI make good decisions, but it may make it a little harder for people to program it. It does this using the total AP force on the map, then it does something else for each of the formations to modify it based on all VISIBLE units within about 9 hexes of the center of the formation. It uses this final multiplier to determine how the AI reacts.

This initial check is the one that I wonder if I shouldn't just eliminate (except for the final push[:D].) I understand why it's there, and how it can help, but it seems like it makes it harder for designers. The other option I might pursue is to treat the designer value as an override instead of a modifier if set to anything other than 'stand'.

I don't understand enough of this to make a useful comment, but the 'VISIBLE units within 9 hexes' thing is something I think Soren Johnson discussed. He said that a human player will generally know the probability of a unit being in a location even if its not visible, and in may cases it is therefore 'ok' to let the PO see units that have been seen previously. I think that's what he said.

Yes, but if done right, this should let Elmer fall for feints, etc. which he won't if he knows where the units actually are (at least that's my thought.) It should make him slightly more aggressive since he can't see all the units.

[ralphtricky]

ORIGINAL: Grognard

Another question would be is the 'overstacking' necessary. I would think defense in depth would be better. What would other players do if facing a human opponent ?

Exactly; this illustrates the difference 'tween Elmer & us. Against Elmer I will overstack to get the odds under 2:1 because I know Elmer won't attack. Against a Hu-man - never, never, never - defense in depth is more tactically sound. Reversing the roles, I would certainly do a frontal assault against a red stack at less than 2:1 just to produce the excess casualties and I would expect any Hu-man opponent to do the same. Whereas if I see your defense in depth (even with > 2:1 odds for the initial attack), a possible counterattack would certainly give me pause in the example above.

Would it be possible to program Elmer with a chance variability in his attack routine? i.e. instead of Elmer just crunching numbers - if certain conditions are met (red stacks, objective hex, key terrain hex) give him a binary yes to attack where he normally wouldn't.

To what extent does Elmer currently recognize key terrain? Chokepoints, flank anchors, etc.

It might be possible to add some more stances (or tune the existing ones) so that Elmer will attack at < 1:1, the problem is that with combats taking a variable number of rounds, 1:1 odds are more likely to burn the whole turn, that's why he doesn't like low odds.

He doesn't recognize chokepoints or flank anchors, he's strictly objective based right now. He does recognize good defensive terrain and bad defensive terrain, and bridging engineers try to bridge rivers, but that's it for right now.

[governato]

ORIGINAL: ralphtrick

This is an example using 3.4. As the German (grey) units are on the defense, they are in a good terrain position (the mountians to the southwest have badlands, so they are good flank protection). But, they are too dense, on minimal losses and not dug in. It seems to be a good defensive position, but with a horrible defensive unit arrangement. Can Elmer make a determination like 'if on defense, and stacked in the red, go for some depth' ?

Probably, one issue is that the only defense that Elmer knows right now is a a rolling defense, not a static one. I'll take a look at what he does on defense so see if I can stiffen it.

My two priorities for AI improvements are connected with the ongoing discussion:

- reduce the amount of unnecessary movement of the AI units when on the defence. That will increase their supply and entrenchment levels.

- improve the ability of Elmer to create a 'continuous' front. This ideally would be a parameter to set by the scenario designer as a function of scenario scale/time period etc.

[Grognard]

He does recognize good defensive terrain and bad defensive terrain

Sometimes at the expense of any semblance of a defensive line - and never any flank protection...

improve the ability of Elmer to create a 'continuous' front

The difference here is contiguous vs continuous. Elmer always makes a straight contiguous line even when there aren't enuf units to accomplish the mission - so he doesn't. This is where we (humans) will alternate hexes if we have to. And reinforce the bends in the line. Elmer doesn't do bends and sometime you just have to. I think this (adequate/sensible defensive lines) would be tricky programming. Sorry Ralph - no pun intended.

[Crking]

I decided to register to see if I could maybe contribute to this thread. I do play a a fair amount of TOAW, but almost exclusively against the PO. Trying not repeat anything anyone else has said (sorry if I am) I would say the greatest weakness of a objective based system is there is no true strategic reserve. For smaller and shorter scenarios this is not a problem but when you get to the large and long scenarios this will leave the PO at a real disadvantage against a player. When a problem or an opportunity appears the player can use their reserves while the PO cannot. You would almost want a group of units without objective that could then take a formations objective (if it is worn out) or share one (to bolster attacks and defence). If the PO could do this, which I know is no simple matter, it would certainly help it. Although once you have the PO being able to do that, it then needs to decide when it should do that.

[Telumar]

ORIGINAL: Chris K

I decided to register to see if I could maybe contribute to this thread. I do play a a fair amount of TOAW, but almost exclusively against the PO. Trying not repeat anything anyone else has said (sorry if I am) I would say the greatest weakness of a objective based system is there is no true strategic reserve. For smaller and shorter scenarios this is not a problem but when you get to the large and long scenarios this will leave the PO at a real disadvantage against a player. When a problem or an opportunity appears the player can use their reserves while the PO cannot. You would almost want a group of units without objective that could then take a formations objective (if it is worn out) or share one (to bolster attacks and defence). If the PO could do this, which I know is no simple matter, it would certainly help it. Although once you have the PO being able to do that, it then needs to decide when it should do that.

This is indeed a PO weakness, but one that can at least partially be compensated by programming a reserve in the editor. In theory the designer can immobilize a formation when under PO control and release it by an event when certain conditions are met like a hex (with or without radius) changing ownership, the variable reaching a certain value, by the victory differential etc pp. However scenario designers can not forsee all possibilities and a flexible, intelligent reserve logic as you propose would certainly be better than a 'hard coded' one.

[secadegas]

ORIGINAL: Telumar

ORIGINAL: Chris K

I decided to register to see if I could maybe contribute to this thread. I do play a a fair amount of TOAW, but almost exclusively against the PO. Trying not repeat anything anyone else has said (sorry if I am) I would say the greatest weakness of a objective based system is there is no true strategic reserve. For smaller and shorter scenarios this is not a problem but when you get to the large and long scenarios this will leave the PO at a real disadvantage against a player. When a problem or an opportunity appears the player can use their reserves while the PO cannot. You would almost want a group of units without objective that could then take a formations objective (if it is worn out) or share one (to bolster attacks and defence). If the PO could do this, which I know is no simple matter, it would certainly help it. Although once you have the PO being able to do that, it then needs to decide when it should do that.

This is indeed a PO weakness, but one that can at least partially be compensated by programming a reserve in the editor. In theory the designer can immobilize a formation when under PO control and release it by an event when certain conditions are met like a hex (with or without radius) changing ownership, the variable reaching a certain value, by the victory differential etc pp. However scenario designers can not forsee all possibilities and a flexible, intelligent reserve logic as you propose would certainly be better than a 'hard coded' one.

Totally agree with Telumar but I'd add something.

PO has obvious shotcomigs but most PO "problems" are clear designer flaws

PO is damned difficult to program. Takes (a lot of) not exciting time and demands (a lot of) expertise.

In 10/+ years of fora experience what i see is that even before understanding TOAW - or even "fire a pbem turn in anger" - there are a lot of "designers" and "scenarios".
And then - because nowadays most people can't lose too much time to go deep in something - a lot of claims about whole system...[8|]

[secadegas]

double post, sorry