Problem in FFT_for_Period(...) in TimesNet.py about time-series-library

[wuhaixu2016]

Hi, thanks for your comments.
(1) It is notable that the i-th element in the FFT results represents the amplitude of frequency-i, which means that indices are frequency and the values are amplitudes of corresponding frequency.
(2) Actually, in our experiments, the top-k frequencies are quite stable. You can run the code for validation.
(3) If seq_len << period, I believe that the first thing you need to do is to enlarge the input length. All the forecasting models will fail in this setting, since the input cannot provide enough information.