A question about hyper-parameter "fftwarmlen"

[zhycs]

Hi, very lucky to read this wonderful paper on periodical modeling and saw your publicly available code.

However, I am confused about some parameters in your code but it is not mentioned in your paper. That is "fftwarmlen" used in the "function warm_PM_parameters_perK" (utils.py - line 100). It is set in different constants varying with datasets, and then determines the length of the FFT handled. Could you tell me how to set its value?

[weifantt]

Thanks for your interest. This parameter "fftwarmlen" is to control a time interval that is used to select most useful periods. Thus, from 0 to (len(train_series) -"fftwarmlen"), it is used to conduct FFT. From (len(train_series) -"fftwarmlen") to len(train_series), it is used to select periods. The different parameter setting of different dataset is by considering the length of the overall series. For example, given a 8-year time series data, you can use 6-year data to init FFT and use the left 2-year data to select periods. Generally, this parameter is not a big deal to final performance; thus you can select a meaningful parameter as you like.