DotA 2 is somewhere between Crack Cocaine and Heroin on the addiction scale. The other side of DotA's learning curve is that apart from being steep, it is also very deep: you can STILL learn so much more at the top skill levels because the game space is so large and unexplored.
I personally ended up building some custom Markov Chain Monte Carlo models (because with the Steam API you can pull a lot of match data) with shared priors just to understand the farming/win relationship per hero (and you need hierarchical Bayesian models because with 100+ heroes, 10 heroes per game, and needing to account for between-hero relationships, you run into the curse of dimensionality quickly).
It's almost more fun to analyze DotA computationally and watch tournaments than to play it (I watch more than I play). That's a truly ridiculous skill space and I could probably do an entire PhD's worth of statistics mining just off DotA 2.
I personally ended up building some custom Markov Chain Monte Carlo models (because with the Steam API you can pull a lot of match data) with shared priors just to understand the farming/win relationship per hero (and you need hierarchical Bayesian models because with 100+ heroes, 10 heroes per game, and needing to account for between-hero relationships, you run into the curse of dimensionality quickly).
It's almost more fun to analyze DotA computationally and watch tournaments than to play it (I watch more than I play). That's a truly ridiculous skill space and I could probably do an entire PhD's worth of statistics mining just off DotA 2.