A data cleaning procedure using the New York Public Library's crowd-sourced restaurant menu datasets.

This reproducible document models the fundamental strength of each national soccer team to ever have played a match. Using a comprehensive dataset of all international soccer matches since 1872, we employ a multilevel negative binomial regression model to asses team strength, controlling for strength of the opponent, the home-field advantage and the team-specific home-field advantage (“stadium effect”). A multilevel approach allows us to “nest” together teams of the same nation but different year, which squares neatly with our intuition that national teams employ similar players, coaches and strategies in nearby years. We model goals scored (and allowed) as a count with negative binomial distribution, a generalization of the Poisson distribution for when the tendency to score (or allow) goals may be heterogeneous across teams.

reproducible report

An easy-to-use and highly accurate text prediction algorithm

Motor Trend Dataset

Four of the Normandy invasion's beaches.

Predicting body movement class given accelerometer data.

Analysis of which storm event types cause the most damage to human health and also economics (property and crop losses).