Exploratory Community

FB Prophet: Same Parameters but Different Results using Python vs Exploratory


I’m using the FB prophet algorithm in Exploratory and recently had our Data Science team complete a model validation using Python. The results from the Python script were different than what I came up with in Exploratory. I confirmed that all parameters were the same but still came up with different results.

How can I do a full comparison of the source code in R (from Exploratory) to Python to ensure that all parameters are the same? (i.e. Fourier order or any other parameters). Can we get access to the source code based on the parameters that we select?

1 Like

Hi Shahab,

The R source code that calls FB’s prophet package functions is open-sourced here.

For the Fourier orders of the seasonalities, for yearly and weekly, we just take the prophet’s default, which are 10 and 3 respectively.

For other seasonalities, we picked 8 and 6 for quarterly and monthly so that they are in line with those prophet default, in that roughly the square of the Fourier order is within the same order as the days in the period.