Recently there have been a number of issues related to seasonalities that depend on other seasonal factors:
@vhpietil provides a solution in #434 for constructing separate daily seasonality features for weekends vs. weekdays, but it's a lot of effort.
We should figure out what a clean interface would look like for a more convenient solution to this problem.
A big thanks to @Olof-Hojvall who added this new functionality in #894 and #873.
There is now an example in https://github.com/facebook/prophet/blob/master/notebooks/seasonality%2C_holiday_effects%2C_and_regressors.ipynb where there are different weekly seasonalities for winter vs. rest of the year (scroll 2/3rds down to the section "Seasonalities that depend on other factors")
If you have a time series with this property, please install the latest github version with this new feature, try it out, and report any issues.
You can get the latest github version like this in R:
devtools::install_github('facebookincubator/prophet', subdir='R')
In Python, remove any existing fbprophet installation (pip3 uninstall fbprophet), clone this repository, cd into the python directory, and run
python3 setup.py install
Pushed to PyPI and CRAN
Most helpful comment
A big thanks to @Olof-Hojvall who added this new functionality in #894 and #873.
There is now an example in https://github.com/facebook/prophet/blob/master/notebooks/seasonality%2C_holiday_effects%2C_and_regressors.ipynb where there are different weekly seasonalities for winter vs. rest of the year (scroll 2/3rds down to the section "Seasonalities that depend on other factors")
If you have a time series with this property, please install the latest github version with this new feature, try it out, and report any issues.
You can get the latest github version like this in R:
In Python, remove any existing fbprophet installation (
pip3 uninstall fbprophet), clone this repository, cd into thepythondirectory, and run