There is a crucial "bug" in the way of calculating metrics.

[zezhishao]

Hi, Thanks for your wonderful works~
Recently, I found there seems to be some inconsistency between STFGNN and other baselines (e.g., Graph WaveNet, DCRNN) in the way of calculating metrics, which may make the comparison unfair.
Specifically, most of the baselines calculate the metrics at horizons 3, 6, and 12, while STFGNN seems to calculate the metrics among horizons 1~3, 1~6, and 1~12.
Do I understand the code correctly?

[EEHITer]

Similarly, I also noticed this problem. The author's first version of Arxiv used an average method to compare the performance of the METR-LA and PEMS-BAY datasets among 1-3 , 1-6, and 1-12 , but works such as DCRNN or GraphWaveNet calculate the metrics at horizons 3, 6, and 12.