Nixtla / neuralforecast

Scalable and user friendly neural :brain: forecasting algorithms.
https://nixtlaverse.nixtla.io/neuralforecast
Apache License 2.0
2.9k stars 332 forks source link

Mismatching NBEATSx parameters: harmonics/seasonality and polynomials/trend #605

Closed rcslight closed 1 year ago

rcslight commented 1 year ago

In the NBEATSx implementation, it reads on the docstrings of the model that:

"""
**Parameters:**
(...)
`n_harmonics`: int, Number of harmonic terms for trend stack type. Note that len(n_harmonics) = len(stack_types). Note that it will only be used if a trend stack is used.
`n_polynomials`: int, Number of polynomial terms for seasonality stack type. Note that len(n_polynomials) = len(stack_types). Note that it will only be used if a seasonality stack is used.
"""

When in fact, harmonics are given to SeasonalityBasis() and polynomials to TrendBasis().

kdgutier commented 1 year ago

Hey @rcslight,

Thanks for bringing this to our attention. The documentation is now improved.