Calculate a weighted RMSE. Weights are based on 'time' in the case of the weights variable. (E.g. older errors weigh less). And also based on the 'direction' e.g. predicting to few used ressources is worse than predicting a few ressources used too much.
weighted_rmse(
actual,
predicted,
weights = exp(-(length(actual):1)/14),
worsenGoodExpections = 1.5
)Real Data, vector of observations
Predicted Data, vector of observations
Time based decay. Default is an exponential decay
Factor by how much predicting too few used ressources should be punished more.
weighted RMSE