For the Textbook's implementation of Seasonal Decompose (STL) with FourierSeries & RidgeCV.
(if I understood correctly)
In .\src\decomposition\seasonal.py
class FourierDecomposition(BaseDecomposition):
...
def _calculate_fourier_terms(self, seasonal_cycle: np.ndarray, max_cycle: int):
"""Calculates Fourier Terms given the seasonal cycle and max_cycle"""
sin_X = np.empty((len(seasonal_cycle), self.n_fourier_terms), dtype="float64")
cos_X = np.empty((len(seasonal_cycle), self.n_fourier_terms), dtype="float64")
for i in range(1, self.n_fourier_terms + 1):
sin_X[:, i - 1] = np.sin((2 * np.pi * seasonal_cycle * i) / max_cycle)
cos_X[:, i - 1] = np.cos((2 * np.pi * seasonal_cycle * i) / max_cycle)
return np.hstack([sin_X, cos_X])
...
def _prepare_X(self, detrended, **seasonality_kwargs):
...
seasonal_cycle = (getattr(date_index, self.seasonality_period)).values
return self._calculate_fourier_terms(
seasonal_cycle, max_cycle=np.max(seasonal_cycle)
)max_cycle=np.max(seasonal_cycle) is wrong
ie,eg:
seasonal_cycle = df.index.dayofweek.values => we know: dayofweek gives 0,1,2,3,4,5,6
max_cycle=np.max(seasonal_cycle) => it takes 6 -- this is wrong
it should take 7 as the max_cycle (-- the period in FourierSeries)
eg:
For a comparison of a simple ex
using 7 clearly fits better than using 6.
eg:
For a comparison of a the textbook energy_consumption data:
(this is not so obvious)
(but you can see, if use 6, there are always 2 days sticks to the same level,, in each period)
For the Textbook's implementation of Seasonal Decompose (STL) with FourierSeries & RidgeCV.
(if I understood correctly)
In
.\src\decomposition\seasonal.pymax_cycle=np.max(seasonal_cycle)is wrongie,eg:
seasonal_cycle = df.index.dayofweek.values=> we know: dayofweek gives 0,1,2,3,4,5,6max_cycle=np.max(seasonal_cycle)=> it takes 6 -- this is wrongit should take
7as themax_cycle(-- the period in FourierSeries)eg:
For a comparison of a simple ex
using 7 clearly fits better than using 6.
eg:
For a comparison of a the textbook energy_consumption data:
(this is not so obvious)
(but you can see, if use 6, there are always 2 days sticks to the same level,, in each period)