DyDo skirmish AI: ver 2.1.1 (21 Apr 2010)

[crux]

game crashes if you forget to install the dydo mod when loading a savegame
anything that can be done to fix this or should this be a bug in trac or is this not a bug?

[Per]

We could call it a 'missing feature', I guess. It should have been done a long time ago, but it is somewhat technically difficult without breaking the savegame format badly.

[Zarel]

I suppose it might be possible to give an error message instead of straight-up crashing, though.

[guciomir]

DD how is your work going?

[DylanDog]

The new version is now available for download on the DyDo website. I understand this is also the version which will replace AIvolution in the next Warzone 2.3beta release. This means you can form that point onwards use the WZ Trac to raise tickets related to DyDo.
Suggestions should be posted here not in the Trac!!!.
Possibly I will close the DyDo website...I am not sure about that now.

[guciomir]

:D:D:D

It is a great idea to incorporate Dydo in official releases

I wonder what do you think about my last long post from previous page?

[DylanDog]

I wonder what do you think about my last long post from previous page?

I answered it already , see here

[KukY]

I finaly tested DyDo after a long time.
I really like way of building on gateways. I doesn't just build on it, but also next to it, so now you actually lose some units when passing through gateways!
And I played a game where there were 2x3 DyDos against me+1 DyDo.
Those 6 DyDos only attacked me, they didn't even fight between themselves!
While I barely survived, my AI ally was able to kill all other AIs, because no one attacked him!

[guciomir]

I wonder what do you think about my last long post from previous page?

I answered it already , see here

you can find 2 more interesting posts below , i think you missed them
viewtopic.php?f=10&t=4107&start=135#p45893
viewtopic.php?f=10&t=4107&start=135#p45996

[DylanDog]

I finaly tested DyDo after a long time.
I really like way of building on gateways. I doesn't just build on it, but also next to it, so now you actually lose some units when passing through gateways!
And I played a game where there were 2x3 DyDos against me+1 DyDo.
Those 6 DyDos only attacked me, they didn't even fight between themselves!
While I barely survived, my AI ally was able to kill all other AIs, because no one attacked him!

Priority 1 target is the strongest enemy player (human or AI)
Priority 2 target is the weakest enemy player (human or AI)
Priority 3 target is related to the type of game (number of teams and players) and to DyDo AI strength, which is set at the very beginning of the game. AI strength has five levels: 1,2,3,4,5
In a 8 players game with 2-3 teams and 1 human player the probability to choose the human player as target is the following:

Prob % = 20+random(20)+AIstrength*10

which means that the probability to attack an human player is between 30% and 90%, depending on the AI strength and the random(20).
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
You can check the AI targets (human or AI) on this way:

In the WZ shortcut to play with DyDo add --debug script --debugfile dydo-debug.txt

My shortcut looks like this:
F:\Warzone2100_2.3beta9\warzone2100.exe --window --configdir "F:\Warzone2100_2.3beta9" --debug script --debugfile dydo-debug.txt --mod_mp dydo-ai.wz

On this way DyDo prints off a large amount of statistics I use for debugging, you can search for the following text:

"Att Human Player" and "Att AI"

you will find few rows like the one below:

script |02:31:08: [scrDebugFile] 55min; Orange #1 [1-1] ;0 ;Pl 0-1: 'DylanDog' #0 [1] 'Orange' #1 [2] 'Grey' #2 [3] 'Black' #3 [4] Teams(4) nEnemyPlayers (3) nHumanPlayers(1) nPlayersInGame(4) Att Human Player(12) Att AI(4) anyOilResLeft(0) nTotEnemyDroids(45)


This means that this AI player 1 (Orange ) after 55min has attacked 12 times a human player and 4 times an AI player.

[DylanDog]

Yes, defending derriks with VTOLs is a good idea.

Implemented in DyDo 2.0

I would also suggest creating an event when Dydo decides to split its defend group into subgroup A and subgroup B. Defend range for "A" is still 45 tiles, while range for "B" is bigger so Dydo defends derricks with this subgroup. I have no idea if it is good solution, what is your opinion?

Could be a good option, I need to think how to implement this.

In one of the attached savegames my artillery was killing Dydo defend group and it did not try to attack my mortars/hovitzers. Vtols should attack my artillery and also what is the purpose of defend group if it defends only vs short range weapons? Maybe defend range should be increased when player builds artillery near Dydo? But in this case it would be easy to lure Dydo into trap. What is your plan here?

In DyDo 2.0 if you attack with DyDo with artillery then it should strike back regardless of the defend range.

- skip flamers but use that research time for machine guns / cannons / rockets . It will make them stronger.

mmm, I am not sure that skipping flamers is a good idea...

More ideas:
1. implement event when Dydo builds cyborg transports, puts infantry inside and then send transports near the edge of the map and attack player base from the behind

Cyborg Transports are in my to do list.

I think this is because Dydo sends attack group right away and its is being constantly "recycled" because units die, units are created. On the other hand defend group sits in Dydo base, does nothing and is becoming bigger and bigger but from the time perspective many units are very old..

That a defend group is becoming bigger and bigger should not happen in DyDo 2.0, when an attack is activated, only a minimum number of units is left to defend the base (around 10).

[Per]

Priority 1 target is the strongest enemy player (human or AI)
Priority 2 target is the weakest enemy player (human or AI)

I once wrote an AI that thought like this, as well. Not for this game, but another. It turned out that it was not very fun to play against. While we work optimize the AIs, we should not forget that the most important quality of an AI not to win, but to give a fun challenge for the player.

[zydonk]

Priority 1 target is the strongest enemy player (human or AI)
Priority 2 target is the weakest enemy player (human or AI)

I once wrote an AI that thought like this, as well. Not for this game, but another. It turned out that it was not very fun to play against. While we work optimize the AIs, we should not forget that the most important quality of an AI not to win, but to give a fun challenge for the player.

Amen to that. A mod for Earth 2160 allowed player to choose between weakest or nearest enemy. WZ Ai invariably chooses nearest only, which can be used against it. I find through experience that attacking weakest/nearest is better option than strongest, because the aim is to reduce opposing firepower, and weakest/nearest is best option here.

I think the AI can be scripted to choose between nearest and weakest based on a threat calculation - or between armed and unarmed bots (AI often uses repair and radar units to draw fire, while the real business moves in behind them), especially in defence. This would make attacks less predictable, a good thing.

btw, DyDo plays really well. The difference is sometimes very subtle. It's great that someone has finally taken the time to do a thorough job on the AI. I look forward to seeing how the project evolves. Modern computers can cope will really complex decision making on the fly, and AI development is one way of making WZ even more unique (?? you know what I mean, I hope).

[DylanDog]

Amen to that. A mod for Earth 2160 allowed player to choose between weakest or nearest enemy. WZ Ai invariably chooses nearest only, which can be used against it. I find through experience that attacking weakest/nearest is better option than strongest, because the aim is to reduce opposing firepower, and weakest/nearest is best option here.

I think you are right but I have to say that looking at the debugging reports this happens already in most of the games DyDo play. What I mean is that in most games there the probability that at some point one player gets weaker is higher then the probability that some player get stronger.

I will work in the direction you suggested anyway as the principle is fine:

P1 - Weakest nearest enemy. {8 players game, 4 players games with 4 teams}
p2 - Strongest enemy
p3 - Random {using the AI strength to calculate the probability to attack an human player}

btw, DyDo plays really well. The difference is sometimes very subtle. It's great that someone has finally taken the time to do a thorough job on the AI. I look forward to seeing how the project evolves. Modern computers can cope will really complex decision making on the fly, and AI development is one way of making WZ even more unique (?? you know what I mean, I hope).

Thanks, what do you mean with "The difference is sometimes very subtle"...my poor english...

[DylanDog]

We should not forget that the most important quality of an AI not to win, but to give a fun challenge for the player.

Well this is what I am aiming for already. I do not think that DyDo is currently too strong for an human player but gives him a good challenge.

In 1vs1 games AI will always loose against a medium WZ player, I need to work a bit to improve DyDo and balance some tasks for 1vs1 games.

In 4/8 players game WZ players should try between the various type of alliances and "find" in this way the best challenging option.

An example...this works for me: if I play in 2vs3vs3 (all Generals 1) where I have only one ally, starting from T1 level I manage to get to T3 level where then my ally AI usually gets too weak and we loose...but if I set enemy AIs with some general2/3/4 then the game is really well balanced.