|
|
zkc
Tavern Dweller
|
posted July 08, 2016 02:32 PM |
|
Edited by zkc at 14:35, 08 Jul 2016.
|
magnomagus said:
Quote:
*I have pointed to you WHAT is wrong and WHY it is wrong.
I perfectly understand what you wrote, I was only pointing out a similar list of weaknesses could be written for the H3 and TOE AIs and none of all 3 AIs can be considered 'malfunctioning' because of their defects. The results will simply vary based on the maps you play with them.
Defect and malfunctioning are synonyms. Malfunctioning of AI is proven by existing on a single map and does not get disproved if is not observed on other maps.
magnomagus said:
Quote:
What you don't understand is I was writing about was that replacing 1 AI that has in its disposition x castles and a vector of resources Y with n allied AI instances each controlling x/n castles and having at its disposal Y/n resources will not produce any more challenge than single instance of AI (save more weak heroes to slay).
You cannot make this kind of assumptions if you don't consider the alignments of the towns and the extra scripted bonuses multiple players get in comparison to a single player.
You misunderstood once again. Replacing 4 AIs with 1 makes sense for my point if all x towns are of same alignment IF we are supposed to maximize challenge. I thought it was obvious. Before you misunderstand again - yes, I am aware that 4 AIs would be able to get just 2 properly aligned heroes each. And no, I never assumed 4 AIs allied vs you are supposed to share alignment. This is getting somewhat bothersome.
magnomagus said:
Quote:
This easily degrades to 1 AI controlling all castles and resources the 4 AIs you imagined have (save more weak heroes to slay and considering map is small enough to be properly/fast enough explored using 8 AI heroes).
That's nonsense and only proves you haven't played the type of game I'm talking about, since the AI will not even take over each other's castles if they are allied.
Which proves again you don't understand my point. What you're combating is ALGORITHM. Running n instances of the same algorithm does not produce more challenge than running SINGLE INSTANCE of the same algorithm while giving it n times more resources.
To avoid another misunderstanding you could produce ... You can pick on what I wrote stating 4 AIs would get 8 sets of startup creatures while 1 AI would get just 2 first week. The answer is - place all AI heroes on map and give them starting armies, deduct suitable amount of gold.
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 08, 2016 02:59 PM |
|
|
I should probably have added to my earlier example that I'm talking about a game like:
1 Human Haven vs 1 AI inferno - 1 AI necropolis - 1 AI stronghold
and all AIs allied.
In light of that example what you said here, is nonsense:
Quote:
This easily degrades to 1 AI controlling all castles and resources the 4 AIs you imagined have (save more weak heroes to slay and considering map is small enough to be properly/fast enough explored using 8 AI heroes).
Also what you said about 1 AI controlling resources/towns of 3 players vs 3 AIs controlling the same amount of resources/towns is purely theoretical, since in practice all Homm AIs start to degrade in performance if the amount of towns they need to control rises.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
Elessar
Adventuring Hero
|
|
posted July 08, 2016 04:11 PM |
|
|
magnomagus said:
Quote:
- AI rushes to kill "wandering stacks", but leaves intact all treasures and artifacts these stacks guard AND ignores some guarded buildings after killing guards (e.g. magic shrines)
only if it doesn't need it, doesn't affect Ai combat strength
I've seen the Ring of Celerity left behind-what hero couldn't use a +10% boost to initiative?
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 08, 2016 04:30 PM |
|
|
|
zkc
Tavern Dweller
|
|
posted July 08, 2016 05:52 PM |
|
Edited by zkc at 22:41, 08 Jul 2016.
|
Let's make something clear: I am using latest version of this mod (RC8, downloaded 3 days ago).
magnomagus said:
I should probably have added to my earlier example that I'm talking about a game like:
1 Human Haven vs 1 AI inferno - 1 AI necropolis - 1 AI stronghold
and all AIs allied.
In light of that example what you said here, is nonsense:
Quote:
This easily degrades to 1 AI controlling all castles and resources the 4 AIs you imagined have (save more weak heroes to slay and considering map is small enough to be properly/fast enough explored using 8 AI heroes).
I will remind you that you pointed multiple AIs is solution to leaving resources and artifacts scattered all over the map by Q-AI. I do not see ANY reasoning that could support that on your part.
Now regarding your claim above - even if SINGLE AI controlled 4 DIFFERENT castles it should still be capable of making proper use of racial specialties IF it was well written. Single AI controlling 4 different towns using 4x more resources should do pretty much the same things on adventure map. Joining armies could be more problematic than in case of 4 identical towns, but well written AI should be able to handle it, because management of conquered castles is an integral part of the gameplay. Single AI could even prioritize some buildings better across the kingdom. The only part where 4 allied AIs don't degrade to 1 is they can't combine their force (it would be nice if they could).
magnomagus said:
Also what you said about 1 AI controlling resources/towns of 3 players vs 3 AIs controlling the same amount of resources/towns is purely theoretical, since in practice all Homm AIs start to degrade in performance if the amount of towns they need to control rises.
Which means these AIs poorly scale up (I don't recall experiencing that in H3 times). This does not falsify what I wrote - if you use n instances of malfunctioning algorithm, you will get n sets of flawed results. There MAY exist some special cases when flaws don't show immediately. Even if you pump map full of heroes of 5 AIs and "solve" the problem of resources and artifacts not being collected, this will not make the Q-AI work as it should, it will just make its shortcomings not show in this particular case.
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 08, 2016 06:10 PM |
|
|
Quote:
I will remind you that you pointed multiple AIs is solution to leaving resources and artifacts scattered all over the map by Q-AI.
No that's not what I said, I only wanted to point out the validity of using the Quantomas AI for playing games as the example above.
(and all ARMG templates for more than 2 players set up according to their parameters should fit that as well)
At the same time you keep theorizing about irrelevant game setups and examples I never brought up:
Quote:
Single AI controlling 4 different towns using 4x more resources should do pretty much the same things on adventure map.
I have never seen any HOMM AI do that properly and I don't care anyway since there are almost no maps or templates in existance that start with such a setup (or result in it if the AIs are allied). Also with such setups the turns would take so long in TOE the game is effectively broken.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
dredknight
Honorable
Supreme Hero
disrupting the moding industry
|
|
posted July 08, 2016 06:43 PM |
|
|
zkc said:
I will remind you that you pointed multiple AIs is solution to leaving resources and artifacts scattered all over the map by Q-AI. I do not see ANY reasoning that could support that on your part.
Now regarding your claim above - even if SINGLE AI controlled 4 DIFFERENT castles it should still be capable of making proper use of racial specialties IF it was well written. Single AI controlling 4 different towns using 4x more resources should do pretty much the same things on adventure map. Joining armies could be more problematic than in case of 4 identical towns, but well written AI should be able to handle it, because management of conquered castles is an integral part of the gameplay. Single AI could even prioritize some buildings better across the kingdom. The only part where 4 allied AIs don't degrade to 1 is they can't combine their force (it would be nice if they could).
not sure if I want to go in this fist fight but ZKC, there is one serious point you dont add to the equation of your facts.
Heroes 5 gameplay architecture is immersively more complicated than what HoMM 3 offers. If you dont have this in mind it creates huge blind spot in your bias when comparing AI behaviour. Here is an example of what I am talking about.
1. Heroes 3 has 4 hero stats but only 2 of them make difference - Attack and defense. So basically a good AI will be one that picks might over magic heroes. Even if you play Magic and the AI is dumb you will still lose on big maps if his hero is might and he has equal army to yours.
2. Heroes 3 magic system is simple
- you got damaging spells (damage output increase is pretty straight forward), which besides implosion and armagedon suck big time after early game. So they are not big deal if the AI uses them
- curses/blesses.There 2 type of curses that rule the game - Haste/slow bless/curse. They are all mass casted if you have expert skill.
- Mind control - Berserk - which does not require AI to be controlled.
- Misc spells - blade barrier, remove obstacle etc.. which I have never ever seen H3 use.
Heroes 5 magic system is far more complicated :
- damaging spell have special effects which can be very good only for those effects even if the damage is low (MAster of fire, Master of Ice, Master of Storms, Deep Freze, In some cases it is better to cast non-empowered armageddon so all enemy creatures get defense reduced which is very complicated to evaluate). Also adding here Locus Swarm and Locus Hive which can really change the game outcome if you know what you do = more sohpisticated AI code.
- Curses and blesses have 2 single mass and AOE cast - this means someone should evaluate the best one to get.
- mind spells - default berserk is the same as h3, H55 berserk is so hard to use, Puppet Master require strict initiative/effect-decay calculation in order to be used nice. Otherwise you may waste your cast which actually quite mana unfriendly.
- Misc spells - again AI never uses them as well.
Battle system:
In heroes 3 I consider it simple near advanced. TUrn based where each unit has 1 act per turn. Initiative is based on speed so it is easy to script AI dependency on how to use the units or defend against them.
The more the speed the stronger the creature usually (with few exclusions like Fortress Tier 3 flying snake and the harpy which actually does not care because of NER and strike and return).
Heroes 5 battle system is hard. some units act more but move less or are very fragile. Other are better utilized as support mages which require AI to have logic in casting vs damage output which again decision should not be biased on just on damage because in the long run can have different effect on the battle.
Heroes 5.5 battle system is beyond complex. Everything is highly balanced so there is no ultimate strategy to use (like in heroes 3 Might heroes). So if one AI is stupid the one easy way to make it competative is to cheat with the army because more army means better PvE, more damage and constitution.
Last words...
Everyone likes heroes 3 because of what it is not because its balance.
When something is balanced there is no ultimate strategy to enforce on the AI to use. So AI should be smart. Smart AI requires enourmous effort so the easiest solution after that is just give it huge army.
If you play h55 for a long time and then go to h3 you will be really bored believe me. Out there every battle is almost the same.
Mass slow, wait, shoot. Repeat.
____________
Join our official discord channel | NCF Utility Beta
|
|
zkc
Tavern Dweller
|
|
posted July 08, 2016 11:42 PM |
|
|
magnomagus said:
No that's not what I said, I only wanted to point out the validity of using the Quantomas AI for playing games as the example above(and all ARMG templates for more than 2 players set up according to their parameters should fit that as well).
Q-AI as it is is fit for beta-testing rather than for serious game. Not because it is bad, but because it is incomplete and thus overall worse than original 3.1 ToE. Gameplay examples above.
magnomagus said:
At the same time you keep theorizing about irrelevant game setups and examples I never brought up:
Quote:
Single AI controlling 4 different towns using 4x more resources should do pretty much the same things on adventure map.
I was not "theorizing". Q-AI failed at hero micromanagement (leaving resources, artifacts and buildings after defeating "wandering stacks" defending those, I am not the only person reporting this). It does not matter how many instances of AI reproduce the same mistakes programmed into it. No matter what the game setup is, except for the outrageous case of map filled with AI heroes. Funny quirk is Q-AI flags mines after it kills guards, but leaves resources where they are ... Obviously mines are scored way above resources in Quantomas adventure map exploration algorithm.
magnomagus said:
I have never seen any HOMM AI do that properly and I don't care anyway since there are almost no maps or templates in existance that start with such a setup (or result in it if the AIs are allied). Also with such setups the turns would take so long in TOE the game is effectively broken.
I was playing self-edited Rise to Power (re-aligned all 5 opposing castles and put under control of single AI, gave it 200K gold and 2K wood bonus). ToE AI came after me on week 3 with decent army. Q-AI gathered 40-50 black dragons and did not come after me for 7 weeks now, no change of its actions seems to be coming. In the meantime I conquered 2 fully built castles AI had losing no single unit.
Regarding time - neither ToE 3.1 AI took much time per turn nor Q-AI did. And I am using C2D@0.998~1.83GHZ, so pretty "dated" hardware.
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 09, 2016 12:14 AM |
|
|
Quote:
Q-AI as it is is fit for beta-testing rather than for serious game. Not because it is bad, but because it is incomplete and thus overall worse than original 3.1 ToE. Gameplay examples above.
That is your personal opinion based on a very excentric way of playing the game. Already since the first release most H5 players have moved to the QAI because it offered them better gameplay.
I have never claimed the bug were the AI rarely leaves stuff on the adventure maps doesn't exist but I have seen dozens of ARMG maps that were 90-98% cleaned up by the AI, so it is not a gamebreaking issue.
Quote:
gave it 200K gold and 2K wood bonus).
if you did that in the defaultstats than you broke the mod and your test was invalid. Furthermore it is pointless to do this anyway, it will not provide a boost to the AI, you only motivate the Ai not to pickup stuff since it no longer needs it.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
zkc
Tavern Dweller
|
|
posted July 09, 2016 12:30 AM |
|
Edited by zkc at 01:01, 09 Jul 2016.
|
Dredknight, this discussion above is no brawl on my part. What you described refers to complexity of combat AI and complexity of AIs hero-building algorithm. I can't deny this. Neither do I deny those things were easier to program in H3. All I pointed out above was:
- Q-AI has serious quirks at adventure map level that make original ToE 3.1 AI an alternative worth consideration (or makes worth consideration outright dropping never-to-be-completed Q-AI)
- combat AI in MMH55 (be it Q-AI or not) fails at handling blade barriers and arcane crystals properly
The conclusion on my part is the Q-AI as it is now in game I tested it in proved even less challenging than original one. Should you like to see saves or map, I can upload it or send via mail.
Last time I played H5 (several years ago) was Warlords (the Crossing). I made same thing - realigned all castles and put under control of single AI in hope of getting more challenge, after playing against 5 allied AIs fell short of my expectations. At least I faced bigger armies, but no noticeable slowing down of the AIs turns occurred.
I have long forgotten what ToE AI did on Warlords, probably nothing really interesting. But funny thing - I can still recall the pains I had fighting some strong H3 hero that had dimension door and was wiping my castles being virtually unstoppable. That was over 10 years ago.
|
|
zkc
Tavern Dweller
|
|
posted July 09, 2016 12:48 AM |
|
Edited by zkc at 01:00, 09 Jul 2016.
|
magnomagus said:
Quote:
gave it 200K gold and 2K wood bonus).
if you did that in the defaultstats than you broke the mod and your test was invalid. Furthermore it is pointless to do this anyway, it will not provide a boost to the AI, you only motivate the Ai not to pickup stuff since it no longer needs it.
So you claim flagging mines is OK, cause Q-AI needs them, but abandoning resources next to these mines is not a quirk. Then IF Q-AI has resources WHAT DOES IT NEED MINES FOR?
Aside from that if Q-AI does not need gold, WHY NOT TAKE TREASURE CHESTS AND GET free XP? Did I motivate Q-AI to not take XP either???? Or to not visit lv2 shrines (to make things clear before they get intentionally uncleared - in game vs Warlock, not Barbarian)?
And for your information - wood and gold extras were put on the ground next to Q-AI main hero as pickable gold and wood stacks, so nothing got additionally "broken" in the already malfunctioning Q-AI.
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 09, 2016 01:07 AM |
|
|
Quote:
So you claim flagging mines is OK, cause AI needs them, but abandoning resources next to these mines is not a quirk.
I never said that, you keep putting words in my mouth. I have never denied any of the bugs in the QAI. I have only tried to explain (since the beginning of this discussion) that it provides good gameplay if you follow the rules of the game and let the scripts regulate the bonuses.
Instead you create very unusual scenarios Q probably never anticipated when programming it begging it to fall apart. As developer of H55 I need to adress the desires of hundreds of different players (some of which are already having trouble beating the QAI on normal level). Therefore I don't like the bad advice you spread around based on your excentric playing style.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
zkc
Tavern Dweller
|
|
posted July 09, 2016 01:53 AM |
|
Edited by zkc at 01:54, 09 Jul 2016.
|
magnomagus said:
Quote:
So you claim flagging mines is OK, cause AI needs them, but abandoning resources next to these mines is not a quirk.
I never said that, you keep putting words in my mouth. I have never denied any of the bugs in the QAI. I have only tried to explain (since the beginning of this discussion) that it provides good gameplay if you follow the rules of the game and let the scripts regulate the bonuses.
Instead you create very unusual scenarios Q probably never anticipated when programming it begging it to fall apart. As developer of H55 I need to adress the desires of hundreds of different players (some of which are already having trouble beating the QAI on normal level). Therefore I don't like the bad advice you spread around based on your excentric playing style.
Oh, so EVEN despite I did not touch "defaultstats" ('if you did that in the defaultstats than you broke the mod and your test was invalid. ') I still broke the mod and my test is still invalid. Oh, so Quantomas never expected his AI would have to govern 5 castles of same alignment, having 200K gold and 2K wood above starting resources or face single human opponent 1 on 1 ... And what are the "rules" I am not following (as a side note - every beta-test has a part consisting of "not following rules", to wipe less obvious bugs ...)? Have those been introduced by Quantomas himself, or developed to pseudo-patch holes in Q-AI algorithm, like spamming multiple AI instances to handle somehow Q-AI not collecting resources and treasures as it should (that's way beyond monkey-wrenching approach tbh)? Let me read the rule book ...
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 09, 2016 09:50 AM |
|
|
Quote:
Oh, so EVEN despite I did not touch "defaultstats" ('if you did that in the defaultstats than you broke the mod and your test was invalid. ') I still broke the mod and my test is still invalid.
I didn't say that, you did this part right since you didn't use defaultstats
Quote:
Oh, so Quantomas never expected his AI would have to govern 5 castles of same alignment, having 200K gold and 2K wood above starting resources
It's unlikely he optimized or tested it for such an unusual starting scenario.
If you have more questions about the AI I recommend waiting until Quantomas becomes available for comment again and post your questions in his thread.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
LarkinVB
Known Hero
|
|
posted July 09, 2016 09:57 AM |
|
Edited by LarkinVB at 11:36, 09 Jul 2016.
|
I would prefer reading about MMH5.5 in this thread and btw I'm using the quantomas AI as it gives me a far better experience than the vanilla one.
He (zkc) can try to contact quantomas here :
http://simsysresearch.com/
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 09, 2016 10:12 AM |
|
|
Note that I said 'until Quantomas becomes available for comment again'
He made it clear many times he is unavailable for comment.
So I'm sure it will only annoy him if he gets contacted through non-homm related website. Also he is not a developer for H5.5, so he should not be contacted in relation to this mod. Only in relation to AI as standalone product.
____________
MMH5.5 Downloads | MMH5.5 Translations | MMH5.5 FAQ
|
|
Galaad
Hero of Order
Li mort as morz, li vif as vis
|
|
posted July 09, 2016 01:24 PM |
|
|
I think zkc would be better off playing h7, I heard it has a world-class "expert-AI", can't rival. 
____________
|
|
ldongder
Adventuring Hero
|
|
posted July 09, 2016 04:05 PM |
|
Edited by ldongder at 16:07, 09 Jul 2016.
|
|
I played RC8 map H55-5p-Ultima, Inferno faction, hero is Zydar, I put hero in town and end the turn , but hero is not able to restore mana after he turn, is this a new bug?
|
|
magnomagus
Admirable
Legendary Hero
modding wizard
|
|
posted July 09, 2016 05:23 PM |
|
|
|
Dredknight
Honorable
Supreme Hero
disrupting the moding industry
|
|
posted July 09, 2016 09:13 PM |
|
|
|
|
|
This Popular Thread is pages long: 
