How can it be that a
xgboost.cv cross-validation operation where n-folds are evaluated is quicker than a single
XGBoostClassifier.fit(X,y) of the xgboost.sklearn API?
I believe this is the answer: https://github.com/dmlc/xgboost/issues/651
the sklearn api uses
n_estimators= 100 as default whereas xgb.train is using
As both refer to the same parameter this could explain the huge difference.