Research Project: Searching for the statistically "Optimal" Opening
When choosing my opening repertoire, I looked at lichess winrate stats and, when I had no strong stylistic feeling, often picked the best performing move. I'm sure I'm not alone in wondering if this manual search was missing some better variation or choosing an entirely impractical opening, potentially leaving some Elo on the table.
As a computer scientist, it got me wondering, would it be possible to calculate the opening repertoire with the highest possible winrate, according to the database? To do this, the algorithm needs to not just greedily choose highest winrate moves, but to look ahead and find variations later on that score even better. So I went ahead and solved this problem, creating a completely new search algorithm to do it.
What I've found has changed my impression of what the data has to say about opening choices, and I thought other chess players might be interested.
I've just published parts I and II, explaining why existing tools don't find the optimal solution, and introducing new concepts to solve the problems. Future posts will finish the algorithm, and talk about what the approach has to say about chess: the openings it finds, the limitations of using observational data, and the conclusions I took away from the project.
You can read part 1 here and part 2 here, and let me know any questions you want answered in my next posts!