I was wondering if there were missing values in the 2014-2017 test data to which predictions are compared. If yes, then I guess these missing site-year pairs are not taken into account when evaluating predictions. If not, is it because there have been observations for all year-site combinations or because you have performed some extrapolation to deduce nests counts from the data that was available?
My guess is that there is no data for every site, and every year in the target, since it is costly to get these data.
There must be some (maybe a lot)
From the linear example, only actual measurements are considered when computing the
AMAPE metric, so I’m guessing the same goes with the target.
Thanks for your answer.
That was my guess too, given the number of missing values in the training set. But I was wondering if only the nest counts had been considered in the test set, or if the adult counts had also been used as a proxy for nest counts when there was no data. I guess not, if it hasn’t been mentioned.