Chapter 2: Overfitting

Dear Sir,

I was wondering how you could find such a good result with a RandomForest Regressor compared to my own results; i finally understood that you let the max_depth parameter to its default value which is None.
I remember an interview with coding project where i had let this parameter to None with along other values (small one liike [2,4,8] and the interviewer told me: "never train a RF with the parameter max_depth set to None, because all your result are meaningless since you overfit far too much".
So i was wondering how do you, in practice, try to avoid overfitting ? I personnally plot a validation Curve in order to vizualize the gap between the mean test and mean train score obtained from cross validation. If for one parameter the step/gap is "too hight" i overfit. But i have no idea of what should be the gap to say i overfit.

Thanks for your help