Mean absolute scaled error (MASE) implementation

[klaudia_lubian]

I would like to understand how the Mean absolute scaled error is implemented in Dataiku. For time series, does it take into account whether there is a seasonality in the data, i.e. are different formules applied for seasonal and non-seasonal data as explained here: Mean absolute scaled error - Wikipedia ?

In general, it would be good to get a proper documentation on all Metrics available in Visual Modelling. 

[AlexT]

Hi @klaudia_lubian ,

I believe you may have received a response to this over a support ticket already:

MASE is the mean absolute error of the test horizon divided by the mean absolute error of a trivial identity model applied to all the data (for all time steps in the training + test data, you compute the absolute difference between the current time step and the time step one horizon ago) => this is why it is not 1.
Division by 0 always leads to a NaN metric.

Thanks,