Very cool!

Hmm, well perhaps it's worth me building some algotithms, I can certainly build a good bot for FFAs and multiplayer. Strat stuff would be kinda boring, but if we were to have more than 1 bot I think a 1v1 bot and a FFA bot would make a lot of sense, seeing as the play styles are so vastly different...

Hmm, well perhaps it's worth me building some algotithms, I can certainly build a good bot for FFAs and multiplayer. Strat stuff would be kinda boring, but if we were to have more than 1 bot I think a 1v1 bot and a FFA bot would make a lot of sense, seeing as the play styles are so vastly different...

An ideal AI would detect if it's in an FFA situation or a 1v1 situation and act accordingly.

An FFA bot is actually something we don't have right now. Since the competition was all 1v1, that's all we have.

I was thinking it might be as simple as doing something like "if we're in an FFA, lower our aggression factor and raise our expansion factor" or something like that.

If you have better ideas on how to make an FFA AI, it would be great!

Oh right, if we can incorperate both into one AI that would be best anyway, because once a FFA goes down to 2 players it's a 1v1 situation and the game switches anyway.

What might be best then is if I work on an FFA code, then I can work with whoever is creating 1v1 code and combine them.

I understand what you mean about "aggression" and "expansion" factors, that's probably what I'll focus it around, identifying which factors to alter as the game carries through.

A couple of things, are we allowed to have the bot "learn"? ie. Store information and/or change it's own code (I'm guessing not)

and are we allowed to have the bot make partially random decisions? (like the current AI does)

A couple of things, are we allowed to have the bot "learn"? ie. Store information and/or change it's own code (I'm guessing not)

Currently there's no place for the AI to store data in between turns. The best you can do now is to have it learn during the development process, then store that learned logic in the code.

and are we allowed to have the bot make partially random decisions? (like the current AI does)

Absolutely, and I think it's a good idea to for bots to be partially influenced by randomness. If there's no randomness, people tend to learn how the AI acts and can exploit it too easily.

Personally I like doing weighted randoms, so the AI is more likely to do the things it thinks are best, but it still has a chance of doing something less good (or even stupid) on rare occasions.

Swish.

the Wunderwaffe AI is really good. It won a human.
https://www.warlight.net/MultiPlayer?GameID=10606890

and another one
https://www.warlight.net/MultiPlayer?GameID=10606899

human lost again
https://www.warlight.net/MultiPlayer?GameID=10607131

this one is going to win
https://www.warlight.net/MultiPlayer?GameID=10607115

Edited 2/25/2016 10:03:53

What is it going to do with the human it won?

@Fizzer:

 settings =>
                        {
                            settings["Fog"] = "NoFog"; //turn off fog so we can see what the AI is doing
                            settings["MaxCardsHold"] = 999; //so AI doesn't have to discard
                            //settings["OrderPriorityCard"] = "none";
                            //settings["OrderDelayCard"] = "none";
                            //settings["BlockadeCard"] = "none";
                            //settings["NumberOfCardsToReceiveEachTurn"] = 1;
                        });

can i set just template id?

can i set just template id?

Yes, the API takes a templateID json parameter, I just forgot to include it in the wrapper for the HumanGameAPI (the BotGameAPI already includes it.)

I just added it to the github repro.

Will other Ais be playable for everyone on warlight, or just the official one?

The primary goal is to have one really good AI. If we wind up with multiple really good ones, we could potentially allow both to be used, but that's a stretch goal.

One thing that I think would be enjoyable is if there were AI's of different standards playing on some of the ladders. Would be interesting to see what rating your AI could achieve.

If you changed the matching system on the ladders to prioritise matching Humans with Humans, and only match Humans with AI's when they are very closely rated, and perhaps limit it so that Humans only play against an AI at most 1 game out of every 3, then I think it would greatly improve the ladders and certainly make them more accessible for beginners.

@Fizzer - thx :)

One thing that I think would be enjoyable is if there were AI's of different standards playing on some of the ladders

+1

With very restricted calculation time it's hard to imagine bot that can match top players(well, i'm not AI expert, but one second for so complex game as warlight sounds very small). I wonder what can creators do if their bots would be allowed to calculate moves for few hours in multiday games.

Join FFA with a bot.
https://www.warlight.net/MultiPlayer?GameID=10607199

and this one
https://www.warlight.net/MultiPlayer?GameID=10607242

Edited 2/25/2016 10:52:56

I agree current AI needs a overhaul because I understand WL used to be mostly WR before, which may explain why AI always attacks neutrals of 2 with 4 in SR... Current AI also has no notion of big bonus or empire, defend it, complete it, break it or leave it free so your teammate can finish it (see ROR for best example).

AI will never be good enough to beat good humans as for example I'm not sure he see how much was deployed by each opponent. He can just not use history function which is very useful. And he can not guess where opponent incomes are.

Btw are AIs for random picks only ? It must be a nightmare to code in manual, especially in team game (but it's more for single team).

Edited 2/25/2016 10:58:36

https://www.warlight.net/MultiPlayer?GameID=10607199 - in this game bot played REALLY bad. I'd like to see log with his bonus evaluation.

http://pastebin.com/sbYrNYhE

Sorry, first lines are missing.

cool that it's open source, need to find some time to look at it

http://pastebin.com/sbYrNYhE

Sorry, first lines are missing.

Pity, I was wondering how high he evaluated +1 bonuses, but they are in missing lines :/

EDIT: Also looking on those 2 games, I can see problem with fixed starting locations (like full distribution or non-random warlords) and no wastelands templates. AI will pick in every game the same places and if undisturbed make the same moves. Current AI although making bad pics, makes it every time different.

Edited 2/25/2016 12:07:11

Pity, I was wondering how high he evaluated +1 bonuses, but they are in missing lines :/

Here is the log for the second game
http://pastebin.com/7vr0GYAZ

Here is the log for the second game
http://pastebin.com/7vr0GYAZ

Thanx. As both games have the same evaluation, here's how it looks:

evaluation		territory name		value
				
145		Sereth-Suczawa-Gurahumora	-373,31
				...
100		Triest				-46,28
99		Fiume				-43,52
95		Wien				-28,67
75		Budapest			27,53
				...
1		B?cs-Bodrog			256,52

So he putted them somewhere in the middle. Deffinatly something, that should be corrected. In my opinion starting point should be ["bonus value"/"no. of territories"] variable, and then modified by other factors.

Edited 2/25/2016 12:45:41