i using r forecast package auto.arima() function, testing against predictable sine wave time series. when run r code on local machine in r studio, different output running same code same source data in azure ml. difference can see azure has older version of forecast package 5.4 whereas have downloaded latest version on local machine 5.9. (interestingly older version in azure ml correctly forecasts future values, newer version predicts attenuating amplitude, incorrect).
my question may know why function's behaviour change between package versions, strikes me strange. or missing here? new both r , azure ml..
i have used function auto.arima many times in both local r (rstudio) , aml, same results. aml deployment not easy debugging, tested code locally first. suspect difference may due different versions of forecast package, or different settings of function auto.arima.
to make apple-to-apple comparison, suggest: 1) use same version of forecast package 2) use same parameter setting 3) use same input data.
