|
|
Gidoza
Famous Hero
|
posted April 13, 2020 09:49 PM |
|
Edited by Gidoza at 21:51, 13 Apr 2020.
|
FirePaladin said: @Gidoza
I already have to think and plan enough in real life (also enough challenges). Do I need this in a game I play to relax? No. Simple as that. I don't want never-ending chains of heroes, etc. which I have to plan for the upcoming in-game month.
Edit: Ok, I understand your suggestions on smaller maps, but on bigger maps (those which I usually play), that's another story.
To your first paragraph - if your relaxation means that the standard of a game is to dumb it down to its lowest possible level, then I could only suggest playing a different game.
Moreover, to your second paragraph - in Heroes III, I almost exclusively played XL maps, and I think it applies just as well. Exception not accepted.
HOWEVER! I -would- completely agree to the existence of Town Portal or Dimension Door on "Adventure"-style maps, where the map design is exclusively single-player intended and there isn't really any other "factions" in the game at play. Town Portal/Dimension Door in these sorts of maps ends up becoming a shortcut and convenience in order to make the adventure less of a drag. This is worlds different from actual competitive play or even any non-competitive standard against an AI opponent in a regular game: any game that has mechanics that caused imbalance in any substantial way or are biased against the AI are not features that should standardly be present in a game. The map size is irrelevant.
Again, just to recap: by all means, if you want to play with Town Portal et al, go ahead and do so. What I am saying however, is...
1. For standard maps, the presence of these should never be the default (default should be off, with an option to turn it on).
2. It is a waste of energy to program an AI to account for game features that are detrimental to the strategic value of the game - especially when it makes the AI programming a hundred times harder and probably still will be hugely biased in favour of the player.
So for the sake of this thread in particular - if Quantomas wants to make an AI that can handle every adventure spell, that's fine and I'd regard it as very forbearing and charitable on his part; for me, however - it comes with the caveat that the time could be better spent focusing on the most vital parts of what makes Heroes III wholesome, and Deus Ex Machina mechanics are not that. I don't want to see an AI that is lacking because human time is limited while irrelevant in-game features are unlimited.
|
|
FirePaladin
Promising
Legendary Hero
DoR Modder
|
posted April 13, 2020 09:58 PM |
|
|
@Gidoza
No, that's not what it means. I still play it on higher difficulties though, but whom in their right mind wants to spend 5-10 minutes each turn just planning every step carefully when you need just to relax a little? And realizing the AI is 2 turns away from the town without TP and you are far away isn't gonna help you relax at all (or even if it's 4 turns away and you are even further).
Dimension Door and Fly are totally another story than TP: with them you can technically reach any place, in comparison with TP, which allows you to return to a certain place. I agree they can be disabled, I rarely play with them.
Adventures maps really don't need DD: you can bypass quest guards, etc.
The map size is still relevant, since you won't chain all 7 other heroes along the entire map length just to give you some troops.
And at least H3 AI knows how to use DD or TP.
And what makes H3 wholesome after all?
|
|
Gidoza
Famous Hero
|
posted April 14, 2020 01:50 AM |
|
|
FirePaladin said: @Gidoza
No, that's not what it means. I still play it on higher difficulties though, but whom in their right mind wants to spend 5-10 minutes each turn just planning every step carefully when you need just to relax a little? And realizing the AI is 2 turns away from the town without TP and you are far away isn't gonna help you relax at all (or even if it's 4 turns away and you are even further).
Dimension Door and Fly are totally another story than TP: with them you can technically reach any place, in comparison with TP, which allows you to return to a certain place. I agree they can be disabled, I rarely play with them.
Adventures maps really don't need DD: you can bypass quest guards, etc.
The map size is still relevant, since you won't chain all 7 other heroes along the entire map length just to give you some troops.
And at least H3 AI knows how to use DD or TP.
And what makes H3 wholesome after all?
I said it before and I'll say it again - if your goal is simply to relax, why dumb down a very deep strategy game when you could just play something else? If you're too far away from your town when the AI is in striking range, then that means you have played badly and deserve to lose your town. I don't know what more there is for me to say there.
What makes H3 wholesome is when it is the best it can be. The toughest challenge possible without making the AI cheat and tripling its gold income. There's plenty of relaxing games that are meaningful that don't involve devolving the game in order to achieve the purpose of relaxation.
|
|
FirePaladin
Promising
Legendary Hero
DoR Modder
|
posted April 14, 2020 02:17 AM |
|
|
@Gidoza
One thing: everyone likes what they like at any game. I just simply like how it plays, the visuals, etc. (it's late and I'm not gonna make a list containing everything I like) and that's why I play it, not some other crap. And what, I have no permission to relax playing a HOMM game how I want? There surely are many more players like this. And it's not like TP was introduced in WoG.
Anyway, there's no point in continuing this discussion: at least to me, right now, you don't seem to understand a game can be played out of many more different reasons. And we won't get to an end either.
And you also seem to be quite bored in real life to be that serious about a simple PC game. Don't take this as an insult, though. I might be wrong.
P.S. The thing with the town was just an example of what could happen if I leave the town to explore without TP; exploring and clearing is also a nice thing to do.
P.S. 2 I do play serious matches too, that doesn't mean I don't know how to play well; I just usually choose to sit back and enjoy a casual game instead.
Edit: All of this means that I prefer to have TP than not to, not that I can't play without. And I never play on difficulties (like easy or normal) where the AI is dumber.
|
|
Gidoza
Famous Hero
|
posted April 14, 2020 02:57 AM |
|
Edited by Gidoza at 12:45, 14 Apr 2020.
|
FirePaladin said: @Gidoza
One thing: everyone likes what they like at any game. I just simply like how it plays, the visuals, etc. (it's late and I'm not gonna make a list containing everything I like) and that's why I play it, not some other crap. And what, I have no permission to relax playing a HOMM game how I want? There surely are many more players like this. And it's not like TP was introduced in WoG.
Anyway, there's no point in continuing this discussion: at least to me, right now, you don't seem to understand a game can be played out of many more different reasons. And we won't get to an end either.
And you also seem to be quite bored in real life to be that serious about a simple PC game. Don't take this as an insult, though. I might be wrong.
P.S. The thing with the town was just an example of what could happen if I leave the town to explore without TP; exploring and clearing is also a nice thing to do.
P.S. 2 I do play serious matches too, that doesn't mean I don't know how to play well; I just usually choose to sit back and enjoy a casual game instead.
Edit: All of this means that I prefer to have TP than not to, not that I can't play without. And I never play on difficulties (like easy or normal) where the AI is dumber.
*Shrug* I can just point out how this discussion relates to this thread in particular: why am I in this thread? Because I want to see an AI who can play well without cheating and completely kick my ass in the process. If enough effort was put into it, that AI could use TP, predict your maneuvers and anticipate your use of TP, defend well against your assaults, and so you'd have to go into a whole new state of planning and consideration in order to beat it. In other words - all your notes about having not to plan routes etc wouldn't matter because you'd have an enemy so skilled that even TP would not save you against such a skilled opponent. The game would cease to have the qualifications for "relaxation" that you want. So - why bother even fighting for the cause when the AI that's coming will defeat that relaxation anyways? If you want to relax and use TP, then just don't fight against a tough AI. Moreover, since an AI skilled in TP is more difficult to program, we can just drop the whole TP route in the first place and be done with it.
Either way, your "relaxation" argumentation doesn't work, and that's why I'm going to fight against it. I already said that anyone can play with TP if they want: it simply doesn't make any sense for the standard game to have it as the standard option, or a very complicated perhaps Deep Blue-level AI with limited manpower and man-hours to be programmed to deal with TP when the use of TP is irrelevant to the depth of the game. That's it. I mean - what else are you expecting from this AI? That's my question. If it isn't something that's there to challenge you as much as possible, I just don't see what the point would be. Is it worth putting hundreds more hours into an AI just because you're infatuated with Town Portal? I don't think so. That's my entire point, and an excellent application of the aforementioned note on perfection - or Occam's Razer - whichever you like.
EDIT: I realize I didn't address your concern/question for why it's worth caring about this. A better metaphor to give for this is that after a hard day's work, one might enjoy just having a beer. Why not? Absolutely. However, just imagine that someone has a decades-old, excellent French wine, and after work, opens it up and...just chugs it as if it were some cheap beer. I suppose I could just not care - because it's your wine, your life, so you can do whatever you want with it. But treating expensive wine as if it's part of a six-pack...is horrifying. Or the trope about putting ketchup on an expensive meal made at a 5-star French restaurant - I expect the chef to throw me out of the restaurant if I try that. Or pretending like the Mona Lisa can be put up on the wall right next to some modern art piece of trash that is a black square with a line through it. It's revolting. All those things relate to one might call having cultural sensitivity. You could just as easily ask why someone cares about food so much (and their life is meaningless), or why care about wine (or their life is meaningless), or why care about art (or their life is meaningless). And yeah - I admit, perhaps it is all meaningless and there's just nothing for me to say; the alternative is that it isn't meaningless, and that I'm confronted with a thing of great beauty and wonder, no how matter how small it is - and to treat it as nothing special...I only have religious language that can possibly describe that: sacrilege, blasphemy, etc. Given the options of shrugging my shoulders before something awesome or ranting and raving like a lunatic over it for whatever ridiculous reason I do, I prefer to be the chef who throws someone out of the restaurant, because I don't find it worth being and refuse to be apathetic about and leave it to nihilism. That just sucks.
|
|
lokiseto
Hired Hero
|
posted April 14, 2020 01:29 PM |
|
|
Teleportation, TownPortal
I prefer a game with a strong AI and fully respect the capabilities of Quantomas in that regard.
Parameters like spells or Government are optional as of 5.5
and I don't get the discussion.
Everyone here is able to tweak or customize a game of 12 years of age to his liking - there is no need of trifle talk.
@Gidoza
Have fun playing the game, if not with an AI, get online and
look there for opponents.
Best regards
|
|
FirePaladin
Promising
Legendary Hero
DoR Modder
|
posted April 14, 2020 01:43 PM |
|
|
@lokiseto
I too, want a game with good AI, but I just expressed that I prefer to have TP (if I remember correctly, I did install an AI enhancement for H5. It might be this one). I didn't think @Gidoza would get so triggered though. And I agree everyone can tweak the game to their liking. People can play without some spells, by me, but not taking into consideration different play-styles (not talking about noobish ones, ofc) and saying they aren't relevant and are dumbing down the game (even though a great percent of players play like that, and the style is not dumbed down at all, only with a preference for TP), isn't nice.
Yeah, he could play online PvP, but the bad thing is that you can't just continue playing where you left off when you want, and this is also a reason I don't play online PvP. And to be honest, enhancing the AI isn't bad, let's not forget the AI actually gets bonus (25%?) resources in H3 and H5 to compensate each week and after battles for H5 at least.
|
|
Quantomas
Responsible
Famous Hero
AI Wizard
|
posted April 14, 2020 07:45 PM |
|
|
@FirePaladin
There is no way around it, a strategy game is a thinking man's game. That's where the fun is for those who appreciate it.
Naturally, you can play HoMM for many different reasons.
But I wonder, how would you react if the AI hounds you mercilessly using TP, blocking your every approach to a town it controls, and making you pay if you come too close? Game after game?
In principle to create a quality non-cheating AI that can do this is years of dedicated work. But the groundwork has already been laid and the infrastructure is there to implement it easily. The same goes for Summon Creatures and Dimension Door.
Gidoza is right in one very important point, this isn't a thread that aims to discuss AI for HoMM in general, or what it is supposed to do, but instead has at its core the idea to create an AI that can make H5 into a game that can offer a true challenge to competitive strategy players.
There aren't many games that can offer this. So, indeed focus matters here.
@Gidoza
It is a privilege to meet someone who does understand what strategy is about. Plus that the HoMM games are special and can be so much more with a fully developed AI. It requires vision and experience to see this. My hope is, that once an AI emerges that is worthy of this aspiration, and people experience what it means to play against such an AI opponent, that more people will understand why it is an effort worth undertaking.
Don't worry that spending work on these adventure map spells will diminish the overall quality of the AI. There is way more than 10,000 hours of work on the enhanced non-cheating AI already done. I had the privilege in the last couple of years to work on AI in general, without the pressure of project milestones and schedules, to research and solve fundamental issues of designing true AI. It's all good.
Trust me when I say spending work on TP isn't an issue. It's comparatively small an effort. Having this feature may one day make people understand why it alters the nature of the strategy game that is Heroes V.
____________
Do you love strategy games? Join us on Discord: discord.gg/JKU6tey
|
|
FirePaladin
Promising
Legendary Hero
DoR Modder
|
posted April 14, 2020 08:43 PM |
|
|
@Quantomas
That wouldn't be a problem though, since humans too can aggressively use TP. To be honest, it's a good idea, making the AI more wholesome than before.
And as I said before, all I've done was just say how I like to play (usually, not always), even if it was somewhat overdone. I don't even really remember how we've got to this.
Edit: Oh, so that's why, Gidoza just went with the thing that I shouldn't play HoMM because it's not a game for relaxing. And I continued to defend my point of view, which is that not only the strategic part is important in HoMM (this doesn't exclude the strategic part as being important).
|
|
Gidoza
Famous Hero
|
posted April 15, 2020 01:24 PM |
|
Edited by Gidoza at 13:27, 15 Apr 2020.
|
FirePaladin said: @Quantomas
That wouldn't be a problem though, since humans too can aggressively use TP. To be honest, it's a good idea, making the AI more wholesome than before.
And as I said before, all I've done was just say how I like to play (usually, not always), even if it was somewhat overdone. I don't even really remember how we've got to this.
Edit: Oh, so that's why, Gidoza just went with the thing that I shouldn't play HoMM because it's not a game for relaxing. And I continued to defend my point of view, which is that not only the strategic part is important in HoMM (this doesn't exclude the strategic part as being important).
I apologize for my framing - it's not about whether TP is there or not, it's about having the potential for the AI to be as strictly powerful as possible and having the deepest strategy experience: I simply wouldn't include TP in that mix.
All I want is to get kicked by the AI. It's true I could find human players, but I've often found that quite difficult and also time-consuming for a variety of reasons...besides, H5.5 is...finnicky with multi-player. The AI is a bit more of a natural fit for a game like HoMM which in the end seems geared to VS AI play or playing with close friends on terms that are easy to organize.
I also like fiddling with the game balance - hopefully the AI might be able to understand whatever changes I made.
@Quantomas
And fair enough that you know best where particular efforts lie in terms of what is programmable for the AI.
I also heard you're working on some other Heroes game of your own making...is that a related project, or something totally independent?
Returning to the AI itself - even though none of us can help with programming, I wonder whether it would be useful to have a checklist for AI logic to make sure all bases are covered? I'm sure I speak for a number people when I say that this is very exciting and it makes one want to jump up and down in participation while wishing one could lend a hand: the ways to help are limited, but what is available would be very rich.
|
|
FirePaladin
Promising
Legendary Hero
DoR Modder
|
posted April 15, 2020 01:31 PM |
|
|
@Gidoza
I understand (and agree). But there's a thing I'm kinda confused about: does the AI really know how to use TP or any other adventure spell (besides Summon Boat in H5)? And, if the AI would know how to use those spells well, wouldn't it make it easier for it to defeat a human player?
In H3 at least there was AI value for troops, so the AI there bought troops based on that, but how does it work in H5? By stats or the same AI value?
|
|
Quantomas
Responsible
Famous Hero
AI Wizard
|
posted April 15, 2020 09:19 PM |
|
|
In principle the upgraded AI doesn't use predefined values for interactions on the adventure map. H3 and H5 vanilla used that, but that approach is limited in so far that skilled human players will always find a way to exploit it. Even if they don't know about the numbers, they will recognize patterns in the behaviour of the AI.
Fixed values for towns are unrealistic. For example consider your main hero is from Haven and you have four towns of your faction under your control. Lets say your main has an Orc castle in striking range this turn and will win this battle. Your main has also a Necro castle in sight in four days travel distance in the opposite direction, possibly just arriving on the turn before a new week's recruitment. If you know additionally that the Orc castle is its owners only, and the Necro has four Necro castles and is your main opponent on the map, your choice will be obvious. Alternatively consider there is a fifth Haven castle that you could add to your empire in reasonable range.
What I am saying is you need a sophisticated lookahead to make informed decisions.
The AI I developed does exactly that. To avoid confusion, don't take the AI in MMH55 as a reference as the game was altered without the AI in mind, not by me, mind you. I have no stake in this mod aside from agreeing to have the next release (dubbed 3.1k) compatible with the 3.1j release they used as a base and help them with a few requested features like fixing broken animations and terrain for the classes they added. But aside from that we are on our own.
Now, if we are going to upgrade and analyze the AI properly, there are a couple of things you need to know. The current line of AI upgrades (up to 3.1j released in 2011) is sort of an ultra-optimized hybrid of conventional systemic AI routines. It is capable of processing magnitudes of order more variants than the H5 vanilla AI (think a factor of 10,000 and more) and with this kind of additional processing power does a much more refined analysis of the interactions that can happen on the map. There is an entire architecture of routines that can do any kind of analysis, and so its easy to add support for spells like Dimension Door and Summon Creatures (TP is already in it, as well as Summon Boat).
It's still a hybrid, though. In many instances it uses fixed values for interactions, for example visiting the Artifact Merchant. But if you look closely at it, these values only establish a preferential order of what to visit. But in a larger strategic context these are insufficient. Consider the example with castles above, acquiring ownership of another castle of your faction has a much higher value, and taking one pivotal castle from your strongest opponent has much higher value still. For this the AI has a sophisticated lookahead with four passes and a lookahead range out of reach for H5 vanilla or H3. It would be easy to make the AI a hundred times faster, if that was the goal.
3.1k and the upcoming 3.1.18 beta will be interesting because it boosts the lookahead substantially (from 7 to 21 turns), has a much more aggressive AI, adds a proper layer to handle everything water related (ships, lighthouses, shipyards, docks, summon boat), plus a couple other major upgrades.
It is even difficult to tell how to test it properly. We will need to make sure that it handles every map object properly, but the real challenge is to figure out whether it gets its grip on the strategic context right. It's absolutely context-dependent.
The other thing you need to know, the entire 3.1x line is outdated from my point of view. I spent the last couple of years researching what an AI needs to do to surpass the skills of a highly skilled competitive human player. This AI (3.2) is no longer a hybrid, it arrives at the values for all interactions based on its usefulness in the given strategic context.
Development-wise H5 is doubly challenging. On the one hand you have the deep strategic lookahead, the AI proper if you so will. And on the other hand you have the myriad features H5 (and most HoMM games) offer. The number of different adventure map locations is already substantial, not to mention the number of factions, heroes, and creatures. However, these all are already handled by 3.1x in one way or another. That is while the AI itself will be replaced wholesale, the hooks for what it needs handling are already there.
So, we will have to figure out initially, what you can do in terms of testing and analyzing the AI that will help with the 3.2 AI as well.
Hopefully, the upcoming beta will be ready soon, so that we have something that we can use as a reference.
I will say something more about what else is on the agenda, once we move on to our own dedicated community space.
____________
Do you love strategy games? Join us on Discord: discord.gg/JKU6tey
|
|
Gidoza
Famous Hero
|
posted April 17, 2020 12:15 AM |
|
|
FirePaladin said: @Gidoza
I understand (and agree). But there's a thing I'm kinda confused about: does the AI really know how to use TP or any other adventure spell (besides Summon Boat in H5)? And, if the AI would know how to use those spells well, wouldn't it make it easier for it to defeat a human player?
In H3 at least there was AI value for troops, so the AI there bought troops based on that, but how does it work in H5? By stats or the same AI value?
While it's true that if the AI knew how to use TP well it would be more of a challenge to the human player, I look at it from a completely different standpoint.
Let's say we have a typical 1vs1 game where there are only two towns on the map. Obviously, how the game plays out will depend on the map size, how far away the towns are from one another, etc. Ostensibly, the goal of the game is either to eliminate your opponent or hold his town for a week before he can recapture any. However, a spell like Town Portal (let's look at it from H3's perspective) mucks things up a little.
In a game without TP, if I lose my town, I figure I have 3 possible options of what to do about it.
1. Charge the opponent's town and take it down, making the game even keel because neither of us have our home any longer.
2. Attempt to re-take my town. I'm probably at a disadvantage due to his having walls and turrets, but it's not impossible. This option would end the game one way or another though.
3. Muck around for an artifact or something and see if the opponent simply leaves, being content with having done me some damage by say - causing my Capitol building to vanish, at which point I take my town back without a fight.
In other words, I at least have a lot of clear options, and none of these options self-evidently puts one side at a greater advantage than the other.
However, if you introduce TP - especially a TP where you can go to any town on the map with at least Advanced Earth Magic - the moment a player with such a TP takes the other player's town, the game is pretty much over. If I try to take his town - he can defend it; if I try to take mine - he can defend it, and also get a new set of reinforcements from his home, guaranteeing my loss. While not guaranteed by any means, a technical "sub-victory goal" of the game is to get TP and Advanced Earth Magic before your opponent - which can come from either a mage guild or from a lucky scroll (where the latter has happened to me many times).
In other words - while you could make an argument that TP might be fair for both sides (h vs h or h vs AI) so long as both players *always* receive TP simultaneously, I think realistically any game where one player can have TP while the other doesn't will always be an unfair game, and it's highly based on luck. Luck should not be the deciding factor in a strategy game.
Luck and random chance are a fun aspect of games - sure. But in the end, we want things to err on the side of Chess-like combat with absolute minimal "luck" in order to maximize the strategy. TP's presence isn't a matter of whether the AI can handle it, it's a matter of TP's distribution's always being intrinsically imbalanced. I don't really see a good way out of this, and as far as I'm concerned, the best path is simply not to bother with it at all.
Quantomas said: Development-wise H5 is doubly challenging. On the one hand you have the deep strategic lookahead, the AI proper if you so will. And on the other hand you have the myriad features H5 (and most HoMM games) offer. The number of different adventure map locations is already substantial, not to mention the number of factions, heroes, and creatures. However, these all are already handled by 3.1x in one way or another. That is while the AI itself will be replaced wholesale, the hooks for what it needs handling are already there.
Well, besides all the different factions and adventure locations on the map, the maps themselves also change. Chess - though a complex and deep game - has the AI advantage of being predictable despite having a huge amount of different strategies. H5 is a nightmare.
Perhaps a change in topic - but when I made maps for StarCraft Brood War, I watched carefully how the AI played and then re-designed many maps to optimize the AI's ability. I suspect something similar will have to occur here to have solid games. Your AI may be amazing, but at the end of the day - for example - even though I like H5.5's new adventure map things, I quite frankly plan to design my H5 maps using mainly Vanilla features and attempting to be fairly boring in variety. I don't plan to include Memory Mentors. I don't plan to make exceedingly complex pathing systems. I definitely plan to remove Mentoring from the game as it makes levelling secondary heroes pointless and that annoys me (not related - just ranting). And how well can the AI handle special objectives? (e.g. Kill a specific neutral group, or find the grail - will it bee-line such things, or be able to go through 6 keymaster tents in a row to achieve a goal? I suspect this isn't a good plan to go on from a design perspective.)
My point is that even if you're the most skilled AI programmer out there and I'd like to see an AI handle EVERYTHING, I'm still going to err on the side of caution and cater map design to the AI, and not pretend that it can handle everything. If maps are made for the AI (while still leaving open lots of potential for variety and such - I love games with 20 towns), the most optimum of results are likely to follow.
|
|
Quantomas
Responsible
Famous Hero
AI Wizard
|
posted April 17, 2020 07:32 PM |
|
|
Good to know that you are prepared to create maps for the purpose of testing the AI. That's naturally the best way to go about it.
We already created a few well designed reference maps back in the day. Maybe these could serve as a starting point?
Indeed H5 is highly complex, which makes developing an AI for it an enormous task. But ... if we truly could have an AI that is competent playing H5, this will be a major milestone. And all the more an amazing experience. I believe that most people can't see it without having experienced it.
Your point regarding memory mentors is sound. In principle I will follow the example of HotA that we treat H5 like a good custodian. But we can make adjustments where it does make sense, or add things like new adventure map locations.
Regarding the special objectives, a few are implemented. It is not difficult to implement them, it's just that it requires additional work, and compared to a lot of other features they are currently on the low-priority list. Naturally the objectives are implemented in such a way that the AI goes about it naturally.
____________
Do you love strategy games? Join us on Discord: discord.gg/JKU6tey
|
|
Gidoza
Famous Hero
|
posted April 18, 2020 11:12 PM |
|
|
Quantomas said: Good to know that you are prepared to create maps for the purpose of testing the AI. That's naturally the best way to go about it.
We already created a few well designed reference maps back in the day. Maybe these could serve as a starting point?
Indeed H5 is highly complex, which makes developing an AI for it an enormous task. But ... if we truly could have an AI that is competent playing H5, this will be a major milestone. And all the more an amazing experience. I believe that most people can't see it without having experienced it.
Your point regarding memory mentors is sound. In principle I will follow the example of HotA that we treat H5 like a good custodian. But we can make adjustments where it does make sense, or add things like new adventure map locations.
Regarding the special objectives, a few are implemented. It is not difficult to implement them, it's just that it requires additional work, and compared to a lot of other features they are currently on the low-priority list. Naturally the objectives are implemented in such a way that the AI goes about it naturally.
I can give a quick example of a map if you like - then I have a rather interesting question.
A map I enjoyed playing in Heroes 3 had a victory condition whereby you had to defeat a particular group of monsters. In order to reach the monsters, you had to get to 6 keymasters' tents (at least theoretically - you have to remove Dimension Door and Fly from the game first...), located all over the map, in each of the main territories of the players plus in two heavily-guarded neutral areas. On top of that, you only had 6 months to complete the goal.
What I loved about this is that this setup kept me on my toes the whole time. I can't just hunker down in my own area, even if I am building up armies, because I need a huge amount of hero stats for the final fight. To get those stats, I need to go fight enemies. By fighting enemies, I lose units - which I need in the final battle. I also need to take side-trips where my typical opponents might not be in order to reach keymasters tents to get to the end. And with the time limit, I need to conjure up enough troops after all that fighting to be able to manage the final battle - where only level 1 spells were allowed.
Obviously, the Heroes 3 AI did not "get" the goal. The layered complexity made any actual victories the AI ever achieved in the end purely accidental, besides the fact that there's no way to witness whether the AI gets to the final goal if I am killed before the 6 month period is up. This game would often involve sacrificing my original town intentionally in order to reach certain places because it was worth it - or to put it another way: there were "bigger fish to fry." Even if it's low priority - do you think the AI in the end could handle this level of complexity?
So anyways, an unrelated question - perhaps four years ago I had ran some tests and offered them to Magnomagus who later confirmed a theory I had, which was that Heroes 5 AIs couldn't function after the inclusion of a 21st town on a map. Just a synopsis of that: up to 20 towns on a map, and everything works fine. Add a 21st town, and the 8th player slot for AI just stops working - it doesn't build anything, it doesn't do anything. Add a 22nd town, and the 7th player AI slot becomes disabled.
So here's the question I never got the answer to really: is the AI dysfunctionality a product of the AI's not being programmed to deal with that many towns on the map, or is this a product of H5's only having a registry to record up to a maximum of 20 towns? The answer could be helpful because it determines what kinds of limits can be placed on map design.
|
|
Quantomas
Responsible
Famous Hero
AI Wizard
|
posted April 19, 2020 06:32 PM |
|
|
The 3.1x AI doesn't look beyond what is behind a borderguard if it does not have the means to unlock the gate. But if it can see that it can obtain the key, it will probe that and what it can do behind the gate. Generally it will fight the monsters that give the most experience and provide access to the largest treasures. So it performs to its best abilities, in principle to optimize the gain obtained, so it always tries to be as fast as possible. However, lookahead is important here. For example, if it could obtain a key in turn 4, then open the gate in turn 8, obtain another key in the region behind the borderguard in turn 12, that is all possible with proper event sequencing (probed in various permutations), but the turns add up. The H3 AI has a lookahead of 3 turns, H5 vanilla of 3.5 turns and the H5 3.1x so far 4 - 7 turns. The upcoming beta for 3.1k will have 21 turns with its new exploration mode. This is a significant upgrade.
But it is probably most valuable on maps with standard victory conditions. Also note that 3.1x doesn't do full coordination between its heroes, e.g. it won't see that a secondary hero can obtain a key in turn 6 and thus could proceed in the same turn past a borderguard, making its way there in advance.
On the other hand the 3.2 AI that will be released later does full coordination, tracks victory states generically, i.e. it would see the chain of events required to win, and has effectively an infinite lookahead. We will need to test what it is truly capable of once we have it.
Regarding the maximum town limits, I did indeed implement a fixed limit for 3.1x, but that is rather high, 128 towns. No reason to limit your maps to 20 towns. If the bug persists, it should be straightforward to track it down.
____________
Do you love strategy games? Join us on Discord: discord.gg/JKU6tey
|
|
Gidoza
Famous Hero
|
posted April 19, 2020 08:25 PM |
|
|
Quantomas said: The 3.1x AI doesn't look beyond what is behind a borderguard if it does not have the means to unlock the gate. But if it can see that it can obtain the key, it will probe that and what it can do behind the gate. Generally it will fight the monsters that give the most experience and provide access to the largest treasures. So it performs to its best abilities, in principle to optimize the gain obtained, so it always tries to be as fast as possible. However, lookahead is important here. For example, if it could obtain a key in turn 4, then open the gate in turn 8, obtain another key in the region behind the borderguard in turn 12, that is all possible with proper event sequencing (probed in various permutations), but the turns add up. The H3 AI has a lookahead of 3 turns, H5 vanilla of 3.5 turns and the H5 3.1x so far 4 - 7 turns. The upcoming beta for 3.1k will have 21 turns with its new exploration mode. This is a significant upgrade.
But it is probably most valuable on maps with standard victory conditions. Also note that 3.1x doesn't do full coordination between its heroes, e.g. it won't see that a secondary hero can obtain a key in turn 6 and thus could proceed in the same turn past a borderguard, making its way there in advance.
On the other hand the 3.2 AI that will be released later does full coordination, tracks victory states generically, i.e. it would see the chain of events required to win, and has effectively an infinite lookahead. We will need to test what it is truly capable of once we have it.
Regarding the maximum town limits, I did indeed implement a fixed limit for 3.1x, but that is rather high, 128 towns. No reason to limit your maps to 20 towns. If the bug persists, it should be straightforward to track it down.
Interesting - so are you saying that the 3.2 AI could handle the complex victory conditions previously described? Or would it still be better to go with straight-forward conditions?
As for the town limits - I'm quite certain that I ran my tests on the 3.1j at the time - it should be a simple matter to test whether the problem still exists seeing as far as I know that nothing has changed in that realm since last I tested.
|
|
Quantomas
Responsible
Famous Hero
AI Wizard
|
posted April 20, 2020 02:24 PM |
|
|
The upcoming beta is for 3.1k, which needs testing first.
____________
Do you love strategy games? Join us on Discord: discord.gg/JKU6tey
|
|
Gidoza
Famous Hero
|
posted April 20, 2020 07:39 PM |
|
|
Quantomas said: The upcoming beta is for 3.1k, which needs testing first.
I guess I was confused as to which AI you were referring to previously.
|
|
Gidoza
Famous Hero
|
posted May 18, 2020 01:06 AM |
|
|
Hey Quantomas,
Been doing some experimenting and I think this is worth pointing out in light of the new AI that's going to be coming.
I'm wondering how the new AI will handle the purchasing of heroes. My question is motivated by the fact that the current hero-purchasing system for the AI is extremely mechanical and doesn't favour the AI any way you spin it. I'll explain what I mean.
Presently, the code for AI hero purchasing has three fields - one for the number of base heroes, one for the number of extra heroes to purchase per town owned, and one for max heroes on the map.
Presently in H5.5 (for example) the AI mostly limits itself to 2 heroes - and this is fine for 2-player games and the like. However, as soon as you get onto bigger maps or more players and such, 2 heroes simply isn't enough to handle the situation and the AI begins to croak because it can't be in enough places at the same time - it loses ground needlessly.
On the other hand, if the formula is changed - for example, to 1 base hero plus 1 per town: while it's true that I might want more heroes the further I expand, 1 per town strikes me as excessive. Unfortunately, the AI will quite ridiculously purchase heroes to its hero cap no matter what the situation - so in this case if the AI has 5 towns, it purchases up to 6 heroes, end of story. It just doesn't need that many. 1 per town with 0 base heroes doesn't work either because a second hero at the start of the game is usually pretty important. Anyways - the result of the excessive heroes is that the AI wastes tons of money and ends up having heroes that do nothing; and if they get killed, the AI immediately purchases more, wastes more money, and gets them killed.
To summarize: the current setup is such that the AI can only handle pretty small maps in a meaningful way. Once you get into meaningful multi-player, team, or large maps, the AI will either lag behind due to lack of heroes to manage the territory, or will lag behind due to hero overpurchasing.
Ideally in my opinion, the optimal formula would be 2 starting heroes plus 0.5 per town (so the AI would get a third hero at a 3rd town, a fourth at a 5th town, and so on). Sadly, while I can type in the value, the game does not seem to recognize it and it simply doesn't work.
All that said - I'm just wondering if there's any kind of additional logic you're planning to remediate this situation. Sometimes I have just a main and a runner - sometimes I end up with 6 or 7 for various reasons, but I'm using human planning and logic.
Lastly, the hero cap - I'm happy to leave that at the maximum of 64 because why not - but the real reason is that if the bar is too low (e.g. 16), an AI can be choked out of purchasing any heroes at all if they lose one and then another player purchases one before they can. This is just a ridiculous situation if you ask me.
Finally after the lastly - the AI sometimes goes over their own town cap for some reason, rarely - but I have no idea why.
In the short term, I have two ideas that could help along the lines of going the 1 base and 1 per town avenue until a better solution is reached; I don't know how to implement these ideas though, so perhaps you could help?
1. Is it possible to change the hiring costs for heroes? That is - currently it's 2000 gold for the first hero and an added 500 gold for each existing hero. If that was changed to 1500 base (for example), that would reduce the costs of all heroes - this would favour the AI over the human player seeing as the AI tends to be the one to over-purchase.
My problem with this idea is that I've looked everywhere and cannot find a field that relates to hero purchase price. Would you happen to know where it is?
2. Give the AI 100 gold per turn per existing hero. The AI could use a break anyways and this is minor - this would help to offset the silly expenses incurred by hiring heroes that aren't needed while ensuring that the necessary heroes that *are* needed are still purchased. What would the code for this kind of script look like?
Anyways that you and have a good day,
-G
|
|
|
|