TCO-DemographicMembership

1. One Hot encoding of nominal features: GENDER, REGISTRATION_ROUTE, REGISTRATION_CONTEXT, MIGRATED_USER_TYPE, PLATFORM_CENTRE, TOD_CENTRE, CONTENT_CENTRE
2. Training three XGBoost estimators, each configured with a different objective function, namely: rank:pairwise, binary:logistic, and reg:linear.

Build: cmake . && make

NOTE 1: estimator parameters were obtained from hyperopt runs using muse_estimator.py script. NOTE 2: XGBoost code was tweaked a bit to allow for: 1) amalgamation, 2) compilation on the TopCoder testing engine, and last but not least 3) std::log and std::exp invocations were forced to use double precision, as it turned out to be a score booster (my guess for the reason for that is the glibc version on TopCoder machines which supposedly predates glibc changes described here: http://developerblog.redhat.com/2015/01/02/improving-math-performance-in-glibc/).