xreg helper functions about fable HOT 5 CLOSED

Implement piecewise as an argument of trend (with knots)

from fable.

@robjhyndman
Should trend() start from 0 or 1? tslm starts from 1.

However in the tidy context we introduce an origin, which is typically set as the first time index. So in that sense, index[1] - origin = 0. Thoughts?

from fable.

0 would be good I think, as then the intercept is the value of the trend at the first observation. It makes no difference to the slope, and most people don't try to interpret the intercept in any case, so changing from 0 to 1 is unlikely to lead to confusion.

from fable.

This question also applies for fourier terms.

In forecast, fourier() starts the index at 1, and in fable it currently starts at 0. So the fourier terms are offset by 1, leading to different coefficients. This can lead to confusion, but I think the resulting fourier matrices are nicer starting at 0. Thoughts?

from fable.

Yes, better starting at 0 I think.

from fable.