posted by Malystryx.GDS,
"I think the next great hurdle is having better player value metrics," said Alan "Nahaz" Bester. "I will know when I have a reasonable player value metric when it correctly values supports." The vast difference in game objectives for supports and cores makes the concept of a value metric all the more difficult, however advancements in the data available could lead to more indepth stat analysis in the future according to Nahaz.
"The single biggest thing we need that we don't have now is a way to measure a player's contribution to the taking of map objectives. We have assists but we don't necessarily see tower assists but that's something for the first time recently that we have a full set of data; all kills, all wards placed, all towers taken down etc.. by time and by map location."
"We see the location of kills, we see the locations of all wards placed, so there are going to be things in the next month or the next year or so that are going to be coming along. The Dota equivalent of the NBA shot chart is not far off."
Example of an NBA shot chart for LeBron James (Image: NBA)For those unfamiliar with the NBA shot chart(pictured above), it is a diagram that displays an NBA players positioning when making shots and their success rate. Given all kill data is readily available, Dota 2 kill charts could essentially explain where a certain team is strongest or weakest in a fight. It could also allow for pre-match chat to focus on a certain player and which areas of the match he most liked to roam or most likely to strike.
This is a feature we can see in football with an overview of where the player made contact with the ball, or in basketball where it tracks the location a player made a shot from and shows their success rate from that position. For a viewer and an analyst this data would be the next revolution in the way we look at Dota 2, and also could add further legitimacy to the concept of teams investing in a coach.
Could we one day have one metric to rule them all?
"I've played with a couple of variants myself, the main ideas that I've tried so far have all failed drastically for the obvious reasons that it's very hard to come up with one measure that values both cores and supports correctly but the number one thing when you look at statistics is that all numbers are flawed, all stats are bullshit, it's just about learning the limitations of a particular stat."
I think any such one metric that you would come up with for Dota would have its limitations, that doesn't mean that it's not useful, it doesn't mean you wouldn't look at it."
Someone who also shares the opinion is datdota's Dota 2 statsman and caster Ben "noxville" Steenhuisen who we asked to comment on the idea of one metric to rule them all.
"I believe that as Dota is currently played, there must exist a single value metric which represents the skill of every single player. This doesn't mean that someone with a skill of 150 is absolutely better at every possible thing in Dota than someone with a skill of 149, but it means that from a holistic perspective, the player with the higher rating is better," said Noxville.
"That said, I don't know if it's absolutely essential that we attempt to create a metric like this - what are it's core underlying aims? Measuring performance? Allocating blame or awards? At the moment I'm not sure if we have absolutely all the right tools and features from the data in Dota 2 to make this type of metric accurate - but we're making progress."
A statistic to evaluate drafting
datDota's Noxville"I personally think that there are a lot more variables that need to be taken into account, I don't think that in Dota 2 there is such a thing as one optimal draft. I think optimal drafts vary a lot across teams, you have to personalise such an engine but is certainly theoretically possible."
We quizzed Noxville on his chess engine and asked his insights into how such a system could work and how would viewers or statisticians be able to appreciate the output.
"So chess is a little different to Dota 2. In chess the number of variables is really small - and as a result we can accurately predict several moves into the future, and we can evaluate those positions to find the most optimal ones for us," explained Noxville.
"Our opponent is doing the same thing in reverse, they're aiming to stop us from accomplishing our goal. In Dota 2 - the number of variables in gameplay are astonishing - and we have such a small amount of top quality data to evaluate individual moves, that the only place this type of analysis makes sense is during the draft: turn-based movement and slow enough for us to analyse."
Our evaluations are thus a bit limited: since we can't say how good or bad 5 specific heroes are against 5 other specific heroes if they've never met, we turn to evaluating subsets of our heroes vs subsets of their heroes (i.e. reducing the problem to a series of solvable solutions)," said Noxville.
In other words given the amount of playable heroes in Dota 2 (currently 108 in all pick and 106 in captains mode), and 10 possible slots they could be drafted into, there are are already over 31 trillion possible hero combinations in CM - not even taking into account the bans.
Therefore, despite the quantity of pro Dota 2 we have witnessed, the instances of the exact same draft will be minimal. In order to make anything of the data we have available, we have to break down the statistics into malleable chunks.
"This reduction can be really dangerous if not done properly - and is the primary danger in this type of analysis. If it's done properly, the ideal outcome is something like a live draft interface where it could statistically suggest the absolute best heroes to pick for yourself, or ban from your opponent; and to assess which team is ahead after the drafting phase."
It seems that we're still just at the tip of the iceberg but the possibilities Nahaz and Noxville put forward mean we may well soon witness an incredible transformation in the way we analyse the game.
TLDR version - stop scrolling
> New data means we can track not only when kills happen but where
> Allows for the possibility for a "Dota 2 kill chart" in the future showing which regions of the map are the most "active" similar to the NBA or football.
> A single metric to determine a player's impact is hard to create due to diff roles of supports and cores
> Nahaz believes such a metric is possible but would have its limitations just like any other stat
> DatDota's Noxville stated that the tools to make such a metric accurate are not there yet but progress is being made
> A live draft interface that could suggest heroes to pick or ban and help determine who is winning the draft is possible according to Noxville.
> Statistics are about to get more awesome, stay tuned.
What player metric stats do you want to see in the future?
Comments