[kevin9327]

add_seasonality validates fourier_order and prior_scale, but not period. A zero or NaN period is accepted, and the failure only shows up later in fit, far from the call that caused it:

m = Prophet()
m.add_seasonality(name="x", period=0, fourier_order=3)   # no error
m.fit(df)
# ZeroDivisionError: division by zero   (fourier_series: (i + 1) / period)

With period=float("nan") the seasonality features are all NaN and fit fails with RuntimeError: Error during optimization! from cmdstan, with nothing pointing at the seasonality.

period is the length of the cycle in days and divides the Fourier frequencies, so it has to be positive.

Change

add_seasonality now raises ValueError('Period must be > 0') next to the existing fourier_order check. The check is written as not period > 0 so it also catches NaN. Nothing is stored when it raises.

Tests

New test_custom_period next to test_custom_fourier_order: period of 0, 0.0, -7 and nan each raise, and the seasonality is not added. It fails before the change and passes after. python/prophet/tests/test_prophet.py: 63 passed, 2 skipped (run from source with the compiled Stan model from the 1.x wheel).

This change was prepared with AI assistance; I reproduced the failures above and ran the tests.