Maximum Absolute Square Error

Alphathon
#1

Hi, our team are confused about this criterion, mostly the absolute square part. We understand it as the absolute value of the difference of true value and predicted value to the power of 2. To be more specific, it looks like this (you can use a markdown cell in notebook to view it):

$MASE(y, \hat y) = max(\sqrt{|y_i - \hat y|})$

But isn’t the absolute operation redundant for this?

#2

I believe it says “Maximum absolute error”, not “Maximum absolute square error”. So probably instead of taking the L2 norm, just the max value from abs(y_true - y_pred).

#3

Oh. You should check Alphien Dashboard then. In the section Selection Criteria, it clearly says Maximum Absolute Square Error (50%). Hopefully this is just a typo.


#4

Does this forum thread answer your question above : https://dashboard.alphien.com/forum/314 ?


Let us know if we can clarify further.


Thanks, Lionel.

#5

Hi, Lionel. I understand the definition of maximum absolute error. The thing is, it is written as Maximum Absolute Square Error on Alphien Dashboard , and that is what confused us. I think the Alphien team should correct this asap. Thanks.

#6

Hi -

You are correct - on the main description there was a square which should be there. We changed it.

Please confirm that’s all good.

Thanks again for picking this up ! Happy coding. Lionel.