Return loss function

1. Mean Square Error (Mean Square Error / Quadratic Loss): 

To ensure that the predicted MSE value is not particularly unusual, since the square of this error amplifying section.

 

2. The average absolute error (Mean Absolute Error):

MAE guaranteed performance predicted value in most cases good (not deliberately pipe outliers), because all errors are weighted according to the same linear scale.

 

3.Huber Loss:

Huber Loss between MSE and MAE, given outlier some extra weight but do not give too much.