by JR Cook - 9 years ago     show comments

Last time we looked into the advanced stats tracked by PVP Live and featured prominently on their Hearthstone Pro League, we had an interview with their director of Data Science. Today I take a theorycrafting approach, and address the key complaints about these metrics. LETS DO THIS!

Descriptive Stats

• Minions Killed
• Damage Done to Enemy Hero
• Mana Utilization
• Blowback (Measure of how the player leverages his health total as a resource, think Life Tap or swinging a weapon at a minion)

When most people look at stats in sports, they want to see metrics of success. They want to see stats that can quantify how good a player or team is. In Gridiron Football you want your pass rusher to have lots of sacks, while in Basketball you want to see a lot of rebounds and blocks on your dominant center‘s statline. In Association Football you love a high-scoring forward, and I guess you can’t go wrong with Home runs in baseball.

The main problem with the HPL stats is that they don’t work this way. When you take a player’s overall stats, you can’t directly judge how good he is by looking at the PVP Live metrics. Sometimes you win by having more Minion Kills, sometimes you win by having less. Sometimes you have to be mana efficient, but sometimes you have to Hero Power and Pass so you don’t overextend your board. Sometimes the best play is to use your health as a resource, but sometimes you need to care about your life total so you can stay out of a combo range.

So what are these stats good for then?

Well, while the PVP Live stats might not be telling you whether a player is good or bad, they tell you which strategies they prefer to implement. A player with a high amount of Minions Killed likely prefers control decks and plays more conservatively. While someone with high DEH is more likely to be an Aggro player that takes chances on races. High Blowback shows clear preference for Weapon Classes and Warlock, while your Mana Utilization might indicate that you prefer a combo deck.

PVP Live is well aware of the nature of their stats, and they have embraced it. They don’t claim that the goal is 100% Mana Utilization, in fact, that’s a big reason they call it “Mana Utilization” and not “Mana Efficiency”.

So is this good, or bad? Is the heavy critique warranted?

If you’ve played enough Hearthstone, you know that different decks have vastly different styles. You know that every turn can be a puzzle of its own, and that often times the “best available play” is largely dictated by what your opponent is drawing and playing. The “best available play” is heavily conditioned on the RNG of your draw, the match-up, and the cumulative effect of all previous turns. At the highest level, there are many spots where there’s only one legitimate line of play, and your personal preference in styles is a moot point that particular turn. Sure, the turns where there’s many possible options and the player instinct kicks in, are what might decide a game. Doing statistical analysis over these key choices might help us determine what the best line of play is, but common Hearthstone intuition tells us that there is going to be a whole lot of noise on this data.

Tracking Deck Archetypes

Personally, I think that looking at overall player stats like this can be a bit wasteful. Classifying which kind of decks and playstyles the pros prefer is something I find really cool, but it is something that’s accomplished much better by actually tracking deck archetypes. Instead of just tracking the classes that are played on each game, perhaps you should instead label their decks. You can be really simplistic and split them in 5 main categories:

• Aggro
• Aggressive Midrange
• Controlling Midrange
• Control
• Combo Based

Actually tracking this, will paint a much clearer picture of what kind of player everyone is. If you tell me that a player has 4.7 Blowback, 23 DEH, and 8.3 MK, I can get an idea of what the player likes to play. The problem is that I need to both understand the stats, and interpret the data. More worryingly, I find it a bit hard for the casual players at home to relate to these stats, as it is not something that has been historically emphasized. If you tell us that the same player plays Control 47% of the time, and Controlling Midrange 26% of the time, that’s something any viewer can both understand,and more importantly, relate to.

So wait, does this mean that Twitch Chat has actually won?

Are the stats just a baseless gimmick?

Not quite. If PVP Live were to track deck archetypes, then their stats become A LOT more meaningful. It’s a bit silly to compare the numbers from a Face Hunter to those of a Freeze Mage. But if you only look at the stats a Pro Player has using a specific deck, then the stats can actually be a more accurate reflection of a pro player’s playstyle. If you dig deeper and go to specific match-ups, then you possibly have a powerful tools to evaluate how a match-up is best played.

Statistical Relevance

So why haven’t they done that yet? Well, in order for statistic analysis to be valid, you need a large body of data. If you only have 3 games logged on Handlock, that’s hardly an accurate reflection of how do you play that particular deck. While the pros probably play tons and tons of games on Ladder every month, PVP Live can only track those games that are publicly streamed. The manpower required to track these stats dictates that only tournament matches are tracked, and this strongly limits the depth that their statistical analysis can have.

The biggest challenge for the PVP Live Data Science team seems to be balancing data volume against how accurate a reflection of the game their stats are. This is bound to be an on-going process, as their data library grows, they will be able to provide more accurate results, and dig into more interesting analysis. I’m confident that the future will bring a vast improvement on how PVP Live’s Hearthstone stats are presented to the public, and how well they are received.

I know I told you guys that this was going to be a 2 part series, but the subject is very interesting to me. So I’ve decided to extend this exploration of Hearthstone Stats to a three-parter.  Next Month, I will look at more possible options of why these stats aren’t connecting with the Hearthstone community, and will propose some different stats that might be more in-tune to how the players at home perceive the game.