View Full Version : Skill/Speed Bias? Ramblings for discussion

02-23-2019, 10:02 AM
Does Data Lie? How is it gathered?

I want random other ramblings and theories to think through, so if you've had something bubbling in your head regarding the state of balance feel free to spill it here. I'm not gonna clal anyone out for biased statements, cause very obviously I'm biased towards aramusha.

So I've been puzzling at the season 8 data a lot lately and trying to figure out what to make of it. Aramusha in particular. Also just to verify that win data doesn't include PVAI right? (cause that would explain everything haha) Some people claim it's his rock steady feat that make him strong (it really helps i'll give it that) others claim his feint into gb is too strong (these are the people who havent yet figured out the perfect back dodge timing) and personally if I were to label it on one thing alone I'd actually pin his winrate on his target switching and surprise tactic. If an Aramusha surprises you you get deleted fast. that's just how it goes. This isn't to talk about that, but rather an Identity Idea I've been noticing after watching lots of varied streamers.

Speaking of, a lot of youtubers and streamers were all taken very much by surprise when his(aramusha's) stats popped up, and there's been a lot of discussion as to why he might be where he is, or how it came to be.

So here are 3 initial statements before i get into my theories/rambling. First Youtubers are generally bigger if they are on PC, same with twitchers. Second Ara definitely performs better on console. and Third these people tend to have the biggest voices I think when it comes to community opinions of heroes. just for example Reavyne may or may not be a good player, but he unarguably reaches hordes of people with his opinions. (reavyne chosen at random just because)

Skill/Speed Bias:
Heroes chosen by high tier players only because they are difficult to initially succeed with but have a steep learning curve that only skilled players can truly see and use.

So I'll try to explain my theory but it is kinda hard to grasp, some of it is rambling, thus I'll only state it outright at the end of the post:

Some of the cast are wildly overlooked by the general player, even the platinum rank and up.
*(which btw is very easy to get into... I dont really play ranked but both times i've gotten ranked from unranked I immediately slotted into high platinum then never touched it again since I ranked myself just to see where I stood) and also every time I've attempted to rank myself I did so exclusively with the aramusha.*
Take raider for example. Raider is really hard for new players to do amazing with, but he still hits like a truck. But you know who I see play the most raiders? the pros. partially cause of his dodge GB, and partially because they are real quick with their feint game and parries and he provokes responses constantly. He also demands very quick speed and manuevering. He FEELS like he commands the field sometimes.
Warden another character that lotsa people play in this case, but who only in the hands of someone who really knows how to force his bash game and manipulate people is he a terror. The best wardens I play against don't rely on his SB, they use it sure, but they employ a series of beautiful feints and superior blocks with the occasional zone tossed in. I pick him up and have straight no idea how the wardens I fought against moved and feinted they way they did.
Conqueror: again just as seen here Conqueror is bad in the general populace, but the more skilled the players are the higher his win rate becomes. This is because newer players try to hard turtle and play him slow. Honestly conqueror is also a fast twitch hero. constantly moving between his guards and bashes doesn't require patience, but rather good quick twitches to confuse the opponents.
Black Prior is gonna be this. I've fought mostly terrible black priors. But I've seen the videos. streamers are gods with this hero, because they know how to time everything perfectly already, and what feints actually work well in what situations. seriously the people that are good with this guy are DANG good, and this is because of his fast flow mechanics and versatility.
Orochi, although a noob and console light spamming favorite, is another one. there are those out there who are gods of the feint and deflect game. and tied in with those feints are the storm rushes that feel like IAI blows. fortunately because noobs abuse his light spam his statistics arent inflated as much.

These heroes and their skilled styles tend to be non newbie friendly (despite the vanguard label on two of them this is the case) so my theory is this. Higher skilled/speed characters attract the pros: These more challenging but interesting characters enable pro skills to truly shine and further develop. Thus despite the face value of pick rates we still get inflated data for some of these heroes because they are what define skill to the skilled players. Also since you can grind to reach diamond fairly easily it felt like for me if you put effort into it reaching platinum probably isnt very hard.

Okay this is coming out as a mouthful so heres the short:
The data is inflated because the best players still are the ones playing these characters whether on console or PC and make up the bulk of the pick rate for these heroes too, not because they are OP necessarily, but because the difficulty associated with playing them is satisfying.

I for one feel the most euphoria in a match, not when I turn the tide of a 1 v 4, but rather when I pull of a beautiful aramusha combo with more than 2 deadly feints. Or when I flow perfectly between two opponents to keep them both completely busy whether I'm getting through their blocks or not. My point being heroes that have difficult but very satisfying when successful kits bring out the skilled players thus inflating the data. then those same players play their other characters and dont do as well because they arent as engaged. I love the Shugoki for example, but abso-freaking-lutely I cant parry as well or mind game as well when I play him because I lose my rythm, my flow, and my speed such that it disengages me.

The characters that don't fit this bill make sense. Shaman for example
Shaman is just fun to play. her feint game and maneuverability is awesome, and the easy at which you can perform her soft feints or bleeds just makes her a blast BUT shes so easy to play this way that she's satisfying at all skill levels, so pros kinda get tired of her.
Cent is this way too. I've seen so many memes of cent players being sad cause they scored a GB near a wall and they know it's gonna be an easy win. point being I'm sure cents are actually more satisfied when they win without his cutscene rather than with.

Thus ends my statistical ramblings and here continues a few final thoughts:

Pick rate is not clear to me, is pick rate totaled once for each player??? or total overall. because in the latter case pros still dominate the boards at all levels. Because: although platinum up is the top 4%, the top 4% probably puts in a ton more hours and thus makes up more than 4% of the pick rate across the board.

So are these statistics based on matches individually? or players individually? Likely it's all done with matches cause that data is easier to gather and automate and doesn't require and linkage to specific players beyond defining the data set from which it stems. This means inflation would be caused by playtime then more so than skill level.

Are these statistics constant and how do they rule out data? IE with how weird matchmaking is there is no way players constantly get matched with people of their rank. IE which 4v4s count towards the platinum only ratings? any match that has at least one platinum player? do you then omit the pick rates for the other 7 people that the player competed against/with since they weren't ranked high enough? that would obviously screw the data up. on the flip side do they just throw out that data and only track the data for matches where all 8 members are platinum? in which case there is also scewed stats because many matches and character picks are just omitted because of randomness of matchmaking

So.... I really don't know what to make of these stats @_@

I realize this is a lot to ask, but if a UBI mod sees this I would love for a word to be given to the devs to allow us to see the methods used for refining the data. I get that might be near "company secret" stuff, but It would be very helpful in understanding what we are looking at. And I sometimes get the feeling the people balancing react to the data without caring to think about the methods that wen into it's collection and refinement.

Final comment. As aramusha my fights are usually very polarized. Either I utterly destroy people, or I meet a select individual who has perfected the back dodge so well that I usually dont even get a single hit off. (for some reason my in born enemy as a Aramusha player is either kensei or centurion. Good kenseis just obliterate me when I'm a musha)

02-23-2019, 10:03 AM
That was a wall...... I applaud anyone who reads all my crap thoroughly @_@

02-23-2019, 12:13 PM
I don't really have anything to comment on regarding your theory. But I can comment on why both raider and Aramusha are doing well in 4's.

Raider is rather simple. He can punish things others can't with his ability to GB mid dodge. He has a chained 600ms unblockable that does 50 damage in a wide area of effect and has a variable input window for a soft feint into stunning tap and a generous window for feinting it on reaction to someone trying to hit him out of it. You combine this with target switching to become a monster. Raider is also excellent mid clear and clearing pikeman. Making him a must pick for breach.

Aramusha has absurd damage/damage punishes. 20 damage 400ms light soft feints. Those can be avoided on back dodge. not many do this or are aware of it. Aramusha because of his infinite combo and his damage/reach is also REALLY good at minion clearing. His very nature works well with target swapping. So despite Aramusha being absolute balls in a duel he does work in 4's.

Warden is decent minion clear with high punishes and an unreactable mix up that people are forced to deal with because he has chase down. His top unblockable is also not bad in a gank/team fight situation. And conq is super hard to punish due to his many option selects, relatively safe counter play from bash on side dodge. And he's practically untouchable in midlane if minions are near by.

I honestly couldn't explain Orochi in 4's. Unless you were talking duels when mentioning him. But I agree. I would love to see behind the curtain on how they do their data.

02-24-2019, 01:24 AM
Puzzling over the stats we show is fine, but (and I apologize for repeating myself): data is not a sole source of making balancing decisions. Competitive play, community feeling, and the personal feelings of the devs on what the game needs are also huge factors.

What you guys end up seeing is several nice-looking assets that we create based on a relatively ugly-looking spreadsheet that can be manipulated in a ton of ways to look at data splits. That spreadsheet is populated by game data automatically, AFAIK.

In other words, these particular data sets are not the only ways the team views data, and we're already talking about four different data sets that say a few different things. Please don't get hung up on winning percentages if your main is unexpectedly strong in one or two categories.

The last group of changes we made were Shugoki, WL, and PK. One of those heroes was mainly changed because the devs simply weren't comfortable with their reliance on OOL. Shugoki was a hero that tons of players in the community were asking for changes to, and PK mains have been letting us know that they could use some assistance for a while now (and are continuing to do so).

We've hinted that the next up will be (at least) LB & Raider. You'll find plenty of people in the community asking or us to look at them, and we didn't particularly need data to make these decisions.

02-24-2019, 01:27 AM
I rrally appreciate that long response ubinsulin. Thank you for taking the time.

02-24-2019, 02:55 AM
We've hinted that the next up will be (at least) LB & Raider. You'll find plenty of people in the community asking or us to look at them, and we didn't particularly need data to make these decisions.

Hmm...I like the "at least" statement here...there is hope that even more than LB and Raider will get some love (maybe Gladiator?)

02-25-2019, 12:26 AM
One day Law Bringer will return to his glory days as the biggest bully in town, right now it's Sadmortem to play him, since Warden and Conq can bully better and BP is flat out a better counter attacker.
I still play him because he has the most powerful memes and i hope that makes him more worth while to play, but when you encounter that warden who knows nothing but bash, it's really miserable to be law bringer.

May i suggest giving lawbringer a defended state that puts his pike's spike forward, a counter gank move? you shoulderbash/charge/tackle into it you get impaled and thrown overhead and onto your back as if exhausted.