Closed mkrech closed 8 months ago
Hey @mkrech, thanks for using utilsforecast. We use the third equation here:
$$\text{SMAPE}= \frac{100}{n} \sum{t=1}^{n} \frac{|F{t}-A{t}|}{|A{t}|+|F_{t}|}$$
without multiplying by 100, so the SMAPE is between 0 and 1 (0% and 100%).
Thank you, that solves my problem.
Hi, thank you for your fantastic library.
Smape shows only half of its value if more than one unique_ID is available.
Perhaps the scale could be divide with 2.
In losses.py ... scale = df[models].abs().add(df[target_col].abs(), axis=0).......div(2)
regards Michael