Hi guys,

with each and every Warzone bot in the past, I have run into a wall when it comes to formulate sound play, given the lack of information due to the simultaneous play of Warzone. However when playing myself, I have an understanding which is apparently only intuitive about the opponents options allowing for some error to be safe in important situations.

I am not looking for a perfect solution but my past solutions either completely avoided this topic or were pretty bad. Everytime I think about it too long, my head starts spinning. I need an idea. How do humans solve something so vastly complex without even understanding what they are doing? Are there known algorithms out there, something like a "minimax for simultaneous games"?

Do you mean with simultanous play, the importance of orders of your own move, or all the possible options of your opponent in any given turn?

I assume the latter.
If the bot is limited to only information of the current turn then its probably impossible.
If we assume you do have the information tho, you could make your bot make X pseudo moves for the opponent.
Then have it make X moves for itself. And have some function to play out each combination.
Then have some evaluation function of how good the outcome of each combination is and then play out the set of orders that was on average the best (or had the lowest "worst" outcome)

This does assume a lot of things ofcourse. But its the first thing i came up with.

Maybe i also just missunderstood the question

Edited 12/24/2023 20:36:17

Is the bot allowed to access history or able to get an idea of what the board looked like on turn 0?

i think the bot can only access information of the current turn.
Atleast i recall Justin saying something along the lines of "each turn a 'new' bot plays"
But it might have changed or i missremember

i think the bot can only access information of the current turn.
Atleast i recall Justin saying something along the lines of "each turn a 'new' bot plays"
But it might have changed or i missremember

That is true for bots running on WarLight itself. But I have my bots running on my computer. They can access the game (with some exceptions which don't really matter) just like a human player.

Do you mean with simultanous play, the importance of orders of your own move, or all the possible options of your opponent in any given turn?

People think very little during their games (usually) but the moves still make quite some sense. I can calculate efficient expansion at strong human level and I can also remove the fog at a relatively reasonable human level with ease. But I just can't deal with the nature of simultaneous moves. Even if I would calculate each and every possible opponent play (which is way too much, but I could narrow it down), then which option to choose?


Then have some evaluation function of how good the outcome of each combination is and then play out the set of orders that was on average the best (or had the lowest "worst" outcome)

I assume you mean here that the opponent plays the perfect counter to whatever the bot plays. This would lead to extremely bad play. I have read this also as an option online but obviously, if we border the opponents bonus in 100 different places, no human would ever assume that the opponent just guesses exactly where and how strong we hit him.

I am not looking for perfection here but just for something which is good.

Edited 12/24/2023 23:45:53

-I assume you mean here that the opponent plays the perfect counter to whatever the bot plays.-

I didnt mean that.
Assuming you have the Bot that can already do all the calculations (expansion,unit placements,efficient attacks, ect) and has the information (income,fog knowledge, cycle, ect).
You could have some adjustable settings that influences how the bot prioritises attacking / defending and expanding and how much the bot prioritises these actions (high attack / defense vs little attack / defefense)
Then you have the bot play out and make those orders for the opponent with these difference settings. This will basicly act as the "prediction" of the opponent.
Then you do the same for your own position.
At this point you would have a set of orders made for the opponent and yourself but created them independendly from each other.
Finaly you run a simulation of every combination and choose which set performed best by whatever metric.
So we only consider general ideas and not pinpoint counters.
So even when fighting in 100 different places, your bot would have its priorities set simply on the logic you added, like a human.

-People think very little during their games (usually) but the moves still make quite some sense.-
well id say that is just humans being humans.
You just automaticly discard a lot of options that requires no thought at all.
Then there is the thing of how lazily someone plays.
Like f.e. if im set on defending i wouldnt have think about "will he attack / defend do a small attack / expand?" Which can give the expression of me thinkign quickly and having a reasonable turn even tho i just said "im gonna defend" and did just that. no thought either.