Data Driven Strategizing

Data Driven Strategizing

I apologize for another obscure post, but this topic is something that’s been on my mind since my last blog post on Dota 2 Matchmaking. I was reading more into the data intensive process behind game develop, and I discovered a group of Dota players who run a website called Dota Metrics (linked at the top of this post).

Naturally, any well designed competitive game is characterized by a plethora of different strategies. Take, for instance, one of the oldest competitive games, chess. There are so many mindgames that can be played due to the depth of the metagame. For each decision that your opponent makes, there are logical responses that you can make, and the combination of these responses is your strategy. The effectiveness of your strategy will be quantified based on the result of the match. While there can be some form of calculation involved in a game like chess, the decisions that are made are still largely based on predictions and gut instinct.

Now, enter a game like Dota 2, a game that has hundreds more dimensions than something like chess, simply because it is played in real time. The more dimensions a game has, the more possibilities for different strategies. One example of a strategic decision players have to make is how they want to “skill” their character. As your character levels up, you can allocate skill points to different abilities your character has. The more skill points you allocate to a certain ability, the more powerful it becomes. There are usually multiple ways you can skill a character, depending on the situation in game, so most inexperienced players generally don’t think about it too much.

Here’s where the team at Dota Metrics comes in. They analyze win rates and other metrics for each type of skill build, and since there are at least 4 skills per character, with over 100 characters in the game, each capping out at 15 skill points total, you can imagine how many different combinations there are to analyze. A new character was released last week and analysis of its metrics can be found here. You can see how in depth the analysis gets, which is mind blowing for something as seemingly silly as a video game. They even break it down based on matchmaking ranking categories to show how the skill factor could affect a particular strategy’s efficacy.

Players who are unfamiliar with a new character can simply look at the statistics to determine which method is the correct way to play the character, or at least, which method has the highest win rate based past trends. It almost takes the decision making factor out of strategic gameplay.  Valve, the company that’s developing Dota 2, has realized how important big data is for the game’s community of theory crafters and developers so they’ve recently implemented many new methods of tracking statistics (most notably, DotaBuff). This type of thinking has many implications outside of the video game world too: any time decision making is a concern, think about how much more accurate decisions could be if they were all driven by data? It’s almost ironic that the folks over at Dota Metrics call themselves theory crafters, when there really is less and less theory involved as the game grows and more data is produced.