Why do you exclude records where 'Patv'<0 from evaluation?

[begovegar]

Hi, we don't understand why you exclude records with ['Patv'] < 0 from the evaluation metrics.
According to the paper, you should exclude records with Patv<0 (so=0) when Wdir>2.5. Both conditions should be met.
Is there anything in the report I am missing?
Thanks,

From https://github.com/PaddlePaddle/PaddleSpatial/blob/main/apps/wpf_baseline_gru/kddcup22-sdwpf-evaluation/paddlepaddle/metrics.py:

invalid_cond = (raw_data['Patv'] < 0) |
((raw_data['Patv'] == 0) & (raw_data['Wspd'] > 2.5)) |
((raw_data['Pab1'] > 89) | (raw_data['Pab2'] > 89) | (raw_data['Pab3'] > 89)) |
((raw_data['Wdir'] < -180) | (raw_data['Wdir'] > 180) | (raw_data['Ndir'] < -720) |
(raw_data['Ndir'] > 720))

indices = np.where(~nan_cond & ~invalid_cond)
prediction = pred[indices]
targets = gt[indices]

[ramber1836]

Negative Patv is invalid, which should be not taken into account as well. Regards, Xinjiang Lu

…

On Fri, May 20, 2022 at 9:04 PM begovegar ***@***.***> wrote: Hi, we don't understand why you exclude records with ['Patv'] < 0 from the evaluation metrics. According to the paper, you should exclude records with Patv<0 (so=0) when Wdir>2.5. Both conditions should be met. Is there anything in the report I am missing? Thanks, From https://github.com/PaddlePaddle/PaddleSpatial/blob/main/apps/wpf_baseline_gru/kddcup22-sdwpf-evaluation/paddlepaddle/metrics.py : invalid_cond = (raw_data['Patv'] < 0) | ((raw_data['Patv'] == 0) & (raw_data['Wspd'] > 2.5)) | ((raw_data['Pab1'] > 89) | (raw_data['Pab2'] > 89) | (raw_data['Pab3'] > 89)) | ((raw_data['Wdir'] < -180) | (raw_data['Wdir'] > 180) | (raw_data['Ndir'] < -720) | (raw_data['Ndir'] > 720)) indices = np.where(~nan_cond & ~invalid_cond) prediction = pred[indices] targets = gt[indices] — Reply to this email directly, view it on GitHub <#88>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEVHY3YD5FQEDHQ2INC6ASLVK6EV3ANCNFSM5WPM6HBQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>