Facebook has expanded its presence in the world of open source by releasing an in-house forecasting tool called Prophet. Available in Python and R programming languages, the latest development is designed to provide automated forecasts that are vital for data scientists and analysts.
Forecasting is critical not just for a company like Facebook but even for a startup or a small enterprise. Thus, there is a need for a forecasting tool like Prophet.
Facebook is using Prophet across many of its applications to produce forecasts for planning and goal setting. Based on an additive model, the open source solution is mainly used for hourly, daily and weekly observations with at least a few months of history. It also has strong multiple “human-scale” seasonalities and works best with daily periodicity with at least one year of historical data.
“With Prophet, you are not stuck with the results of a completely automatic procedure if the forecast is not satisfactory — an analyst with no training in timing in time series methods can improve or tweak forecasts using a variety of easily interpretable parameters,” Facebook engineers Sean J Taylor and Ben Letham write in a detailed note.
Facebook’s Core Data Science team is maintaining the code of Prophet. As the Prophet procedure is in Python and R, there is the same underlying Stan code. Developers, however, are free to use the language on the basis of their comfortable level.