Last year a few data scientists including myself exposed severe issues with facebook prophet resulting in facebook pulling down claims such as “with facebook prophet anyone can produce better forecasts than human experts “.
Still the hope was that with Neural Prophet things will be better and one can at least partially repair the worst forecasting model of the XXIst century - Facebook Prophet. valeman.medium.com/benchmarking-f…
My own very limited tests indicated that Neural Prophet was somewhat better than Neural Prophet. It was still “curve fitter” but including AR terms delivered some improvement however the deep AR part didn’t seem to bring any value beyond linear terms.
The large scale examination by @nixtlainc team has shattered remaining hopes - as its parent Facebook Prophet, Neural Prophet seems to be severely handicapped by design.
The mechanistic “curve fitter” design coupled with Bayesian approach seems to put it at severe disadvantage compared even to simple methods invented in the late 1950s whilst being over x100 more slow - a shocker.