Implement a quality metric for recommendations with implicit preference based on the expected percentile ranking.
Map string IDs to unique contiguous numeric IDs that are required by certain machine learning libraries.
San Francisco Crime Classification using XGBoost.
Greedy forward feature selection performing badly on the specific problem.
Improving the SVM classification algorithm by running a grid search optimisation of the parameters.
First try on the Data Science London + Scikit-learn competition.