I love data. A lot.

As a testament to my love for data, my dissertation opens with a quote from The Adventures of Sherlock Holmes (Sir Arthur Conan Doyle, 1892):

“Data! Data! Data!” he cried impatiently. “I can’t make bricks without clay.”

Furthermore, I thrive on competition.

As a result of the statements above, I thoroughly enjoy participating in data science competitions. The leading platform for data science competitions is . Unfortunately, I do not have the time to participate in Kaggle competitions on a regular basis. For one competition in particular, however, I did not manage to resist the urge to seriously get involved. This competition was entitled “”.

The “Finding Elo” competition took place between October 2014 and March 2015 and asked participants to estimate the skill of chess players (as measured through ) on the basis of their moves in a single game. I particularly liked this competition, because it relied more heavily on feature engineering than on large-scale ensembling and meta-ensembling (stacking).

Despite the fact that I am by no means a chess expert, I finished in after six months of intense competition. The most predictive features in the ensemble of models underlying my best submission were based on move sequences, chess engine evaluations, and temporospatial properties of board states and piece movements. I was obviously hoping to finish two places higher, but I guess you can't always beat every PhD in nuclear physics or Harvard graduate turned embarrassingly successful hedge fund manager. Well played, and !

I hope to find some time to participate in another Kaggle competition in the not too distant future.