# does xgboost's eval_metric changes the loss function being optimized?

I'm using xgboost with the reg:logistic objective. As far as I understand, that means that I'm trying to optimize the log-likelihood loss function of my model, and that each boosting step would be in the gradient direction of this function (w.r.t all the examples I've given).

Does changing the eval_parameter to auc affect any of the above? Or is it just affecting what's measured on my validation set examples every 10 steps of boosting?

Chaging eval_metric does not affect the fitted model in any way. You have to interact with objective to do this.