Closed ellisp closed 7 years ago
This might be a recent change of behavior in forecast
. Some fixes were added to auto.arima
and/or ets
for short time series. We can drop the test since it works now, and really I don't know why we were testing this anyway. My guess is it was included as a mistake.
This is really a separate issue/bug that I came across while thinking about it, but check this out:
> forecast(hybridModel(ts(1), models = "ae"))
Fitting the auto.arima model
Fitting the ets model
Error in rep(modelResults$weights[includedModels], times = nrow(fits)) :
invalid 'times' argument
Enforcing a minimum series length (perhaps 4) in hybridModel()
could make sense.
Our tests expect an error in this situation but forecast::forecast does return a (uninteresting) forecast. I'm inclined to remove the test.