Auto Resolve vs. Auto Tactical vs. Human tactical

Let’s see how much the results change based on how we fight this battle against a bone ogre:

image

 

 

Auto Resolve results:

image

 

Next: Into Tactical and just hitting the auto-play:

image

 

Next: I personally play the tactical battle to see if I can do better

image

Now, I burned a ton of mana in my battle but suffered no damage at all.

 

So the question is, why did auto-resolve do so poorly? We’ll have to evaluate that as we go on.

22,133 views 10 replies
Reply #1 Top

Beautiful case study you have here. Did you try a couple auto-resolves also to see if you got different results, just in case there was a critical hit or something else that may have thrown off your results a bit?

Reply #2 Top

How much mana was burned in the auto resolves? Is the AI set to conserve mana in auto resolves? burn thorough it or use it  conservatively? Since that was the first thing you mentioned I would look at that. (although its usually not the obvious answer it doesn't hurt to start there)

But if its using significantly less mana I would say limit yourself to whatever its using in autoresolve and see how well you do.

Reply #3 Top

Can the Auto Resolve accurately model a board setting?

I think modelling distance and space/time may be one of the hardest things to do, but I could be wrong.

 

--> Meanwhile in Autoplay its fought on the board, so the issue doesn't come up

Reply #4 Top


Why is auto-play different from auto-resolve? Is it too much work (cpu cycles I mean) to make auto-resolve simply auto-play the battle in the background, thus always generating the same result?

Reply #5 Top

A few discrepencies between the three results:

1. In the Auto-resolve it gave out 111 XP where the others gave out 110. Is that because one unit died?

2. In Tactical with auto-play there was no Maul awarded while the other 2 got it as loot. Is this awarded randomly?

Reply #6 Top

Quoting Lord, reply 5
In Tactical with auto-play there was no Maul awarded while the other 2 got it as loot. Is this awarded randomly?
End of Lord's quote

I believe the maul actually did get awarded in the auto-play, though it isn't displayed properly.  It's a bug I've seen since beta 2 (I never reported it, my bad) where the loot doesn't properly display in the victory screen, yet is still awarded, so long as the first "slot" is highlighted yellow, as it is in all three SS's.  The bug, obviously, is that the maul isn't properly shown in the middle SS.

Reply #7 Top

To the Original Post.

Fantastic work! I was going to try and poke around this, but these parts are cursed difficult to test around :)

I think one thing you could do to help the computer play in combat is upgrade the AI, ATM I still think AI decisions are terrible.
I think the auto resolve is outside the range of reason though, and with such poor autoresolve I would like to know the outcome beforehand :D

Sincerely
~ Kongdej

Reply #8 Top


Auto resolve does not take into account unit positions. If you look at the details screen it will show units moving towards each other, but at no point in the battle will they stop when reaching other units, they simply pass straight through them to their target (which is always the unit with the least defense). Secondary is that ranged units do not attempt to kite in auto-resolve, but do attempt to kite in auto-play.

The solution to this would be to seperate melee units and ranged units before attempting the calculation. Ranged units always attack the units with the lowest hp/defense and always use their turns to move further away. Each melee unit is capable of blocking up to 3 other opposing melee. Up to 3 melee attacks can be made on a single target, starting with the melee target with the lowest defense/hp. If one side is outnumbered by 2 times or more only one attack can be made against any one target each round.

For every movement speed higher than 2, that unit counts as an additional 1/2 unit for the purposes of blocking to a maximum of 3 (thus making units with move speeds of 6 or more require one unit each to block). Units that are not blocked can make up to 3 additional attacks (total of 6) against any one target and recieve a higher priority for any present ranged units.

To put this in perspective, Let's say an army has 3 archers and 3 melee against another army with 3 melee and 3 (3 move) horsemen. Each sides units make their moves and the archers fire at the lowest defense units. The melee then meet in combat. Since there are 3 melee on side 1 they are capable of blocking a total of 9 units. The value of the opposition against them is 7.5, and thus they can prevent all of the enemies from reaching their rear line. However since they are outnumbered they must spread out to accomplish this and can only strike each opposing unit a maximum of 1 times. Side 2 is not subject to this restriction, and can make 3 attacks against each melee target (starting with the lowest hp/defense). The situation may change if units die on either side.

If in the above case, side 1 instead had only 2 melee, units could break past their lines starting with the fastest/lowest attack units first. In this case a horseman (probably) (worth 1.5) would break into the rear lines returning the value to 6 and enabling the melee to block the rest. Depending on the offensive/defensive value of the horseman (compared to other units on the battlefield) the archers may all switch targets to that unit.

To provide another example. Lets say you attack a shrill lord with your melee champion and a ranged champion. In auto-resolve that ranged champion is almost certain to die, but if you played out that battle you would likely walk away unscathed. The reason for using the lowest attack/highest speed unit is that if that shrill lord (while faster) was even looking in the direction of your ranged champion, you would have your melee champion intercept and let the slower less threatening shrills bypass your zone of control. There probably needs to be some weights attached to the value of a units attack/defense and hp value to properly work this out, but even just making melee stop to fight other melee until they die would be a step up from the current method of caculation.

Reply #9 Top

When the AI fights monsters does it use auto-resolve? That could be a problem for the AI...

Another question: how does the AI check if it should attack an enemy? Is it based just on some computed value, or does it simulate the fights? I guess simulated fights are not cheating as long as fight simulations are possible with the information available to the AI...

Reply #10 Top

Does the AI place any weight to withdrawing wounded units to allow fresh units to take blows?  If the AI never lets units survive (or use spells such as  heal/wellspring) to level up, it is handicapping itself both tactically and strategically.  Especially with some of those nice traits we can design into units. 

This could even add to the faction differentation:  Resoln throws troops at an enemy but gets mana from their deaths (or zombies?)...

Magnar does the same and gets a small reduction in city unrest (nobody wants to get sent to the front!)

Kingdoms or life magic sovs might get special bonuses for a particular hero or unit getting to the brink of death and surviving.  A Battle-hardened troop would be immune to fear/daze effects for example. 

Karavox would also seem to fall into a more troop-conserving category as well.  Why else use shields? 

Do the Trogs eschew shields as for the weak?  or just missile weapons?