An hour later, the Wunderwaffe bot still hasn't made their first move...

Game ID?

Game ID?

https://www.warlight.net/MultiPlayer?GameID=10618719

https://www.warlight.net/MultiPlayer?GameID=10618968

Perfect example why something like bonus:neutrals needs to be added

Adana (South most pick) should have been 6th. :/


You usually pick your best bonuses as 1st.

Different difficulty levels would be handled by passing a flag to an AI telling it what difficulty to play as. In that sense, there's only one AI, it just alters its behavior by the flag that's passed in.

Could you tell us more about flags?

The might be helpful for things like Diplomacies, where idealy you just want an AI to defend it's own territory and not attack out. Can anything be passed to the AI through flags?

If it's been said somewhere else then I missed it, sorry!

Problem with the WarLight AI challenge 2 was that the picking system differed from WarLight. The player there picked alternating pick after pick. Although I always had some problems with a good bonus evaluation Wunderwaffe should be quite aware of income / neutrals ratio however it's Fizzer's replacement code currently which is doing the picks.

Edited 2/27/2016 00:01:57

Don't worry Norman, I think they are only being overly critical because they think they are helping...

Once people start working on new bots they will move on to critisising them instead :P

I'm not complaining about criticism but just explained that some easy to fix stuff doesn't originate from my initial AI.

It would be nice if the AIs figured out their orders during the time that the human players are entering their orders. In other words, the AI should work in the background all the time instead of waiting for the humans to finish and hit commit. Especially for single-player games with lots of AI players, it can take a long time after you hit commit before the turn advances, even on a good desktop computer.

I also noticed that the AI is fairly resource greedy - everything else running on my computer grinds to a halt while the AI is thinking.

If you are planning on new single-players levels on big maps with lots of AI players, low-priority background computations would be nice.

It would be nice if the AIs figured out their orders during the time that the human players are entering their orders.

The problem is that Flash does not support threading. All Flash code runs on the UI thread. It's still theoretically possible to do this by offloading to javascript worker threads, or by breaking the AI work into small chunks that won't interrupt the UI thread for long, but these are all very technically challenging and way out of scope for this project.

Fizzer, could you tell us more about what flags we can pass to the AI?

Fleecemaster, I guess Fizzer can not answer your question because we have not got any AI yet. Someone should implement options before you can choose them with flags. Wunderwaffe is the AI to fight bots in a very specific settings. It does not play well in almost any other settings.

Edited 2/27/2016 21:54:25

I just checked with a ton of debug output why Wunderwaffe keeps picking the wrong bonuses / expanding into the wrong bonuses. Priorising the expansion properly was never it's strength however he wasn't stupid enough to priorise a wastelanded bonus highly. Problem is that WarLight needs a more complex data structure than the AI challenge, for example with a territory being part of multiple bonuses instead of only one. For this reason the transformation from the old AI challenge Java bot to the new C# WarLight bot is quite error prone.

Yeah, I think there are going to be a lot of issues arrising from trying to add in super bonus algorithms.

I fixed the bug with the erroneous bonus prioritization however had to also delete some useful related stuff in the process. You can find the changes in the github repository I just created:
https://github.com/Norman1/WarLightAI

I created 3 games on common maps and with open seats. The AI won 2 of those games.

https://www.warlight.net/MultiPlayer?GameID=10627347
--> Strange that the AI doesn't go more aggressive into the opponent bonuses.
--> Turn 8 it's strange that the bot accepts getting pushed out of the Scandinavian Peninsula.
--> The bot is supposed to priorize Xinjang high since it borders multiple opponent bonuses.

https://www.warlight.net/MultiPlayer?GameID=10627518
--> Very strange expansion turn 1, this doesn't seem right.

https://www.warlight.net/MultiPlayer?GameID=10627558
--> Strange expansion turn 4
--> Again the AI doesn't go aggressive towards the opponent bonuses turn 7 however the reason is here that the opponent is threatening a more important bonus.

I'm not sure about some of those fighting issues however the mentioned expansion mistakes shouldn't have been part of the initial AI. Apart from that this looks now more like the bot from the AI challenge. Of course the AI makes way more mistakes than I mentioned however those are mistakes which it always made.

Edit: Trolls Europe map
https://www.warlight.net/MultiPlayer?GameID=10627998
The AI still fails on "strange" settings like an increasec neutral count however it's already capable to defeat standard non prerequisite 1v1 game joiners regularly on "standard" settings.

Edited 2/28/2016 05:43:05

@Norman

Your AI needs to learn how to finish a bonus = fast as possible....
My game with it on EU Troll's Map: https://www.warlight.net/MultiPlayer?GameID=10627953

From this, am I getting that the plan is to work on Wunderwaffle rather than start from scratch?

You can really do whatever, just keep in mind it will take a good amount of time to get an AI as good as Wunderwaffe is now, so if you start from scratch you have to keep that in mind. That being said, if you have a strategy that's significantly different than Wunderwaffe, then you might want to make your own AI. Personally I'm planning on using my PyBot instead of Wunderwaffe, because I've already spent a lot of time developing it, and I think it could get much better if I improve its heuristic function.

The only thing I worry about is if we all work on different AIs, we'll end up doing more work for the same result. Unless you're just intending to make an AI for just yourself, of course...

@MasterTurtle:
Yeah, I saw that, in fact it was an accident to start the game with 4 neutrals. I believe it didn't even try to take a bonus but just randomly walked around.

From this, am I getting that the plan is to work on Wunderwaffle rather than start from scratch?
I recommend at least copying the data structure.