How to interpret log loss metric?

evaluation_metric
logloss

#1

I know that in log loss, you check the predicted probability against the actual probabilities. But how do you interpret this metric?

  • Does lower value mean better predictions?
  • Is there anything which has to kept in mind when using this metric?
  • What are the pros and cons when using this metric?
  • And is there any tips or tricks or hacks which I should know about?