XGBoost scoring function doesn't match training and evaluation

Question

My training loss (logloss, for validation and training) doesn\'t match the score when I use score(). It is a binary classification (Y is a 1d-array with 0 or 1)