And those two aren’t MOBAs.
Comparing Hearthstone to Heroes of the Storm in how a skill-evaluating program could/would work…
That’s a good laugh before I’m leaving. Thanks.
1 Like
From AI perspective, SC2 is a lot more complex compared to HOTS.
Those are single player games.
Team games are another story.
1 Like
They aren’t, you are evaluating decisions of one player at a time, other players are just independent actors. BTW that’s why self driving cars work 
And evaluating that satisfied multiple interests isn’t really much harder and that’s how self-driving cars will work once majority of them are self-driving and can communicate.
But from skill and play-value perspective, HotS is the harder.
Making a system which truly understands who plays the best is like making a System in SC which can understand which unit did the most for the victory. It’s really complicated, and programming is super hard jfyi.
1 Like
Just because it’s technically possible doesn’t mean that’s a realistic expectation though. I’m sure that with enough time, funding and expertise you could eventually come up with an A.I. that can make a decent assessment of performance in a HotS game. But then the more important question becomes: Is it worth it? How much time and money would it cost to actually develop such an engine? Is the possibility of a slightly more accurate MMR worth investing millions upon millions of dollars?
Depends. In terms of difficulty to play (optimally), I might agree. In terms of determining (and learning) ‘player skill’, I suspect Starcraft might still be a lot easier than HotS though. Complex as it might be to play, the basic ‘goals’ of the game are actually pretty straightforward: Both sides start off equal. Gather resources to build the best army, outfight and outvalue your opponent until all their stuff is dead. You can even track a ‘resource score’ to calculate how efficiently you’re managing your empire.
HotS doesn’t really have any of that. There is no fixed ‘score’ to determine how well you’re doing. There is no equal start. There isn’t even a fixed start, each comp has its own quirks. Different teams are better at different things, and might scale at different rates (Some comps simply get better in lategame, so being ‘behind’ early game means nothing). There is no calculable ‘optimal’ play, as almost everything is team-dependent, even the simple things: Contest Tribute, or soak to have advantage on the next one? Both plays are valid if the team agrees, and both are terrible if your team does the other thing.
Also don’t forget that one of the big A.I. learning tools is simply trying stuff and seeing what wins. This works in a single player game, as there every outcome is a direct result of your personal actions. In team games like HotS this doesn’t work, as your team also affects your outcome. Perfectly played games might still end up in defeat if the enemy outplayed the rest of your team. Mediocre performances might still end up as wins if your team succesfully picks up the slack.
I want to like this 1000 times more lol. I’ve felt this way ever since the beginning. You’ll constantly have people nagging you or calling you a n00b knowing they are going to make it on the mvp screen by doing this or that but you’re the one forcing their team to deal with you instead of objectives or the rest of your team pushing a keep.
I love pvp so knowing when you’ve got your opponent’s attention or you’ve terrified them gives you a hell of an advantage. There aren’t metrics for focus fire, area denial, sabotaging the other team, etc. Would be nice if we had a more in-depth metric system and everyone had their primary contribution highlighted but I imagine that would be complicated.
1 Like
It’s nowhere near millions upon millions. I mean, Deepmind did it as a little PR side project. Again, I am not advocating for anything to be developed, it’s a dead game, who cares. But if I were at the beginning of a path, I bet an investment into a decent matchmaking system would have a much better payoff than a few months of esports sponsorship.
Both sides start off equal. Gather resources to build the best army, outfight and outvalue your opponent until all their stuff is dead.
HotS doesn’t really have any of that. There is no fixed ‘score’ to determine how well you’re doing. There is no equal start…
Different teams are better at different things, and might scale at different rates (Some comps simply get better in lategame, so being ‘behind’ early game means nothing).
Not sure what you mean by equal start. The battlegrounds are symmetrical in HOTS not in SC2. There are three races in SC2 right from the start each with a unique set of units and abilities, just like there are different characters in HOTS. There are upgrades for different units, just like there are builds in HOTS. Your comp in HOTS is fixed, your comp in SC2 consists of 200 characters and they can be completely different throughout the game. In SC2 economic factors vastly depend on which race you are playing and the unit composition.
There is no calculable ‘optimal’ play, as almost everything is team-dependent, even the simple things: Contest Tribute, or soak to have advantage on the next one? Both plays are valid if the team agrees, and both are terrible if your team does the other thing.
Of course there is, it will just depend on what your teammates are doing at a time.
Also don’t forget that one of the big A.I. learning tools is simply trying stuff and seeing what wins. This works in a single player game, as there every outcome is a direct result of your personal actions. In team games like HotS this doesn’t work, as your team also affects your outcome. Perfectly played games might still end up in defeat if the enemy outplayed the rest of your team. Mediocre performances might still end up as wins if your team succesfully picks up the slack.
Again, self driving cars is a “game” with a lot of “actors” and works fine and that’s exactly what the system can do - identify who “picked up the slack”.
Actually, none of that was related. Target has an entire database of what expecting mothers are willing to buy that are not directly baby related. They do this because they want to send them baby catalogs.
When the father’s daughter was being sent all these baby catalogs by Target, the father came in to complain to the manager. The manager then called a few days later, where the Father apologized because the daughter told him she was pregnant.
Yeah, I remember that story. But I think it’s more about datamining, which pretty much everybody does these days whether we like it or not. They never revealed what exactly she bought.
Right, but that’s a side project from a company dedicated to building better A.I.s, and that already had years of expertise and millions of dollars spent learning how to build such A.I.s. (And still ignores how much time and money that ‘side project’ cost them). That’s not exactly representative of how easy it would be for a ‘normal’ (game) company to ship out a full-fletched A.I. like that. Just because an astronaut could (relatively) easily put an object into space while they’re there does not mean that it’s easy for anyone to put objects in orbit.
So why are you even in this discussion? If you genuinely think this is a dead game, leave. If you don’t, don’t waste everyone’s time spouting this dumb trolling.
And how is that relevant for this question? We’re not discussing HotS development strategy here, we’re discussing the viability of the ‘ideal performance evaluator’ now.
I assume you’re trolling at this point, but they are symmetrical in SC2 too.
I meant that each side starts with equal resources, equal space, equal means to adapt. It’s just you and your opponent, no other factors. Any move he makes, you can (try to) counter, and vice versa. There are a lot of different moves (unit types, builds, army compositions) one can make, but at the core it is an equal 1v1 fight, may the best man win. In HotS this is not the case, as you’re dealing with 5 different opponents and 4 different allies, each of which could be of a different skill level. Even when you can outplay one opponent in a direct duel, that’s meaningless in the bigger 5v5 and all accompanying interactions.
Though I admit I ignored the differences in races (Never really played SCII outside of Co-op missions), HotS again seems a lot more complicated. In SCII there’s only 3x3 = 9 different matchups you’d need to find an optimal strategy for. In HotS there’s 85 Heroes, and that’s not even going into the deeper effects of team compositions. Even if you’re playing the same Hero against the same enemies, the kind of support you can get from your own team comp can change your role dramatically. Plays that are amazing in one comp can be foolish in another.
Sure, but how would you rate individual players from that? Most of the time the correct play is ‘do what the team wants’, and that’s meaningless if you’re trying to judge individual skill.
You run into similar problems with micro skills. While there are certainly some solo actions you can perform, a lot of the time you’re reliant on your team to be effective. Whether someone is doing ‘good’ or ‘bad’ can depend entirely on how well their team plays with them, regardless of how skillful their own performance is. A Teams (willingness to) follow up determines if an engaging Tank is getting a brilliant pick or just dying for overextending. Peels and heals determine if a DPS is doing damage or feeding. And so on, and so on.
I admit I’m not up to date with those tech developments, but I thought there were still plenty of problems, mostly with in interactions with other people. And besides, a self-driving car is more like a HotS Bot A.I., which seems a lot easier than such a performance evaluator anyway. Figuring out the optimal way to play the game is a lot simpler than accurately judging how well someone plays.
Blizz have all “BIG data” know when REAL players vote really good players in MVP screen… it won’t be so hard to teach AI to have good results
That’s not exactly representative of how easy it would be for a ‘normal’ (game) company to ship out a full-fletched A.I
That’s why outsorcing exists. Plus, hiring a guy who knows what to do isn’t exactly unheard of.
So why are you even in this discussion? If you genuinely think this is a dead game, leave. If you don’t, don’t waste everyone’s time spouting this dumb trolling.
Why not? You are clueless as to how this problem can even be approached and you are here. If anything I am a lot more aware about possible solutions.
I admit I’m not up to date with those tech developments, …
See, I could address each of the comments in detail, but it’s not going to do any good and not worth the effort.
But then again it’s a lot easier to average the MMR and add some stats together. Too bad the game dies with that approach.
This is a team game. If one side lost, it’s because the members of that team didn’t work together well enough. They deserve the same punishment.
False.
You can be a miracle player, yet you can lose if one ally decides to throw it. And you can’t do anything with that if the enemy has at least 1 working brain cell.
(or unless the enemy has same throwing person)
Ah, the “bad things only happen to me and no one else”.
Your statement doesn’t make any sense…
Stats alone are not a good parameter to evaluate a player. The MVP screen is pretty pointless by itself.
You cannot calculate in numbers a good peeling or a perfect CC chain or a wombo combo setup (just to name a few things that cannot be transfered into numbers).
1 Like
Note that AI uses the APM that are above human speed. Stalker control that could not be achieved by a human.
In other words, the AI might as well be a cheater.