search

antoinecarme / pyaf

PyAF is an Open Source Python library for Automatic Time Series Forecasting built on top of popular pydata modules.
BSD 3-Clause "New" or "Revised" License
459 stars 72 forks source link

PyAF does not detect complex seasonal patterns #73

Closed antoinecarme closed 6 years ago

antoinecarme commented 6 years ago

When playing with some R forecast package examples, the series taylor which has two seasonal periods (Seasonal Periods: 48 336) is not analyzed properly.

PyAF uses two modes : in fast mode, only frequently used/occurring seasonal patterns are analyzed, in slow mode however there should be a deeper analysis (all cycle lengths are tested) which does not seem to be the case.

Forcing more cycle lengths makes PyAF detect the large cycle (336), 336 being a multiple of 48 , this is OK.

Slow mode needs to be corrected.

antoinecarme commented 6 years ago

the option cForecastEngine.mOptions.mCycleLengths can now be set to None to force deeper exploration of seasonalities. Is slow mode, this is the default behavior.

Fixed.