Hypothetical German 1 Move

@nuno:

@Frood:

I guess you could make the AI base it’s calculations on No Luck outcomes

That would be a poor AI.

Agree.
It is not important on what kind of loss forecasting AI is based, it is important to have a solid planning module that may allow the AI to prepare and follow a soundly global plan/strategy.

Good point, I guess. Loss forecasting is only important in deciding where to attack and move/place your forces within the context of a single turn. The main challenge for the AI (at least the biggest issue for the present TripleA AI) is logistics and purchasing. The AI just keeps building Transports and never brings them back.

I think so. The loss forecasting may be used to assign units to the battles but the decision of where to attack should be made with a planning strategy.
In TripleA it is not possible to make exhaustive search in the problem space, using fast forward planning techinques or backtracking strategies. So the only choice, IMHO, is to have high level strategic, logistic and purchasing modules that uses heuristic rules to define sounding strategies, different for each nations. For example, TRN buying may be “penalized” as action for Germany and Russia, have a decreasing bonus based on the TRN buyed for the other nations. I mean bonus and penalization to weight the buying actions.

Another interesting point of AI, is the possibility to have the AI learning in each games, so adapting to the playing style and strategies of the player. Maybe one day someone will start this work on TripleA…

Did it suit like a glove Imperious Leader?

no but i think my comment is like ‘iron in the glove’ to you.

@Romulus:

I think so. The loss forecasting may be used to assign units to the battles but the decision of where to attack should be made with a planning strategy.
In TripleA it is not possible to make exhaustive search in the problem space, using fast forward planning techinques or backtracking strategies. So the only choice, IMHO, is to have high level strategic, logistic and purchasing modules that uses heuristic rules to define sounding strategies, different for each nations. For example, TRN buying may be “penalized” as action for Germany and Russia, have a decreasing bonus based on the TRN buyed for the other nations. I mean bonus and penalization to weight the buying actions.

Another interesting point of AI, is the possibility to have the AI learning in each games, so adapting to the playing style and strategies of the player. Maybe one day someone will start this work on TripleA…

I guess you’d want to set certain objectives, such as “Hold West Russia” or “Hold Egypt”, and territories with ICs should be defined as “defend at all costs”

You seem to know a lot more about AI than I do, but here are some of the principles I would build in:

Planning phase:

Review board position and set certain “waypoints” - eg. should the US be sending units via Afria or Norway? Subsequent steps of the turn will be planned to advance units along this route. Identify where units are needed at the end of this turn, 1 turn from now, and maybe 2 turns from now.

Part of this step would be to assess the defence of key territories: review the current defence of any territory with an IC in it, and the strength of any possible attacks against it, including 2-stage attacks (eg. what happens if UK attacks WE with everything, and US follows with another all-out attack?) - identify how many more Inf etc. needed to defend. If there are extra, these can be tagged as “surplus” to start moving toward the front.

Essentially, this step determines where units are needed, and sets priorities on these needs.

Purchase phase

1st, decide what to build in mainland ICs.
Then, with “offshore” ICs, figure out how many transports are in range, how many land units are already available for transport (allowing for what may be necessary to defend against potential attacks). Then purchase units / TRNs to maximize the number of land units that can be delivered next turn.

Combat

• identify territories that can be “traded” - taken lightly with air support

• identify territories that can be taken solidly, safe from counter-attack

• identify “cherries” - lone transports in range of air power, ICs without AAs in reach of SBRs, empty territories that can be tank-captured and evacuated.

• determine whether trading units is advantageous or not - do you have the piece-count to support a war of attrition, or do you need to preserve units to wait for a better opportunity to strike?

• calculate likely economic results of all potential battles, and value of units that would be left vulnerable afterwards. Find the most economically advantageous combination of battles.

• Probably also some consideration to tactical issues, such as blocking tank blitzes etc.

Non-combat movement / placement

• review priorities identified at start of turn in light of combat results. move / place units to optimize balance between defending vital territories and advancing pressure against the enemy.

Just some thoughts, not very complete though.

Yes, Frood, very good proposal.
I know a little about AI, because I am starting to study Planning techniques, NEural Network and Genetic algorithm for m work. I am still studying and learning so I have still much to do!

Your idea is interestiong because allow the AI to focus the evaluation on particular area. The power of the PC is the possibility to weigh and compare multiple options weighing. The more interestin thing is that different “abstract startegies” may be prepared, then selected randomly and finally “concretized” with optimization techinques.
Conmcretrized means that an abstract strategy, a set of guidelines, is then completed with concrete purchasing choices, movements and combats to perform.

Really, Frood you have give to me an idea that I will try to implement in my runnig project.
One day, when I have more free time I will try to define the design of this AI.

I look forward to it! Let me know if I can help.

@Frood:

Why would you want Jen to mail you a drunk flaming rubber chicken in a bikini?

Hell, why WOULDN’T you?

Well, some people are just weird.  :roll:

@Lucifer:

@newpaintbrush:

I can write a fairly good AI for Axis and Allies.  I can’t do it for chess tho.

Then write the new AI for tripleA!

Payz0rz me?

@Frood:

I look forward to it! Let me know if I can help.

Yes, I have to work in this field, and it sould be really great if I can join the utility with the fun!