mean column-wise mean absolute error (MCMAE)

link: range: [0, infinity)

summary

• A mean absolute error is calculated for each of the n target variables and the score is the average of those n Mean absolute error (MAE) values.
• This is NOT the mean of all the targets
  • Here’s a counterexample:
    • we have two target columns: [1,3] and [99, 200, 500]
    • MCMAE:
    • $\frac{1}{2}(\frac{1+3}{2}+\frac{99+200+500}{3})=134.166666667$
    • mean of all targets
    • $\frac{1+3+99+200+500}{5}=160.6$

pros

Cons