On Mon, 19 Oct 2009, Marco Dieckhoff wrote: > I have a problem, a saved session will not replay what I did manually. > > My step-by-step approach: > > Open CSV File (Colon, Point) > Yes, use as time series > Time Series > > Options: Yeary, Quarterly, Monthly, Weekly, Daily (5, 6, 7), Hourly, Decadic > (translated from german, may not be exact, also on following terms). > > I'm missing "daily (1)"?
What would "daily(1)" mean? Daily 5, 6 and 7 mean, respectively, that there are 5, 6 or 7 daily observations per week. So I suppose "daily(1)" would mean that you have one daily observation per week. I would call that weekly data (which gretl supports). > After importing, I do > Model - Time Series - ARIMA > temp as dependent, hour as non-dependend variable > > All other settings on standard: AR 1, I 0, MA 1, Seasonals all on 0. > Use constant, do not show iteration details, do Hesse matrix, no > X-12-Arima, exact maximum likelihood > > OK => > function calculations: 91, gradient: 37, ... > > In this window, I use the menu to: > Analysis, Prognosis, add 6 > automatic prognosis > plot previous 24 > OK > > And I get the graph I want to see. > > So far, so good. > > Now I save this session in gretl with File - Session - Save Session. > I open the resulting file in WinRar and get the .INP file. > > # Log started 2009-10-19 08:57 > # recording of session commands. please not this must most likely > # be edited before executing The English version of the note above is: "Record of session commands. Please note that this will likely require editing if it is to be run as a script." In other words, we cannot in general guarantee that the command log will function as a correct gretl script as it stands. What we do "guarantee" (in the sense that it's a bug if this doesn't happen) is that on re-opening the session file the state of gretl is re-established as of the time that the session was saved. > I get the error > ? fcasterr 2:01 2:06 > Observation Number out of boundaries (Beobachtungsnummer au�erhalb der > Grenzen) In this instance the problem is one that, admittedly, we should have been able to solve. If you open your session file you'll see that the observations extend to 2:06 (since 6 observations were added for forecasting purposes), but the addition of these observations did not get recorded in the command log. The fix is to insert: dataset addobs 6 before the "fcasterr" command. I'll add a fix shortly, to ensure that when observations are added implicitly in the context of forecasting, this gets recorded in the command log. Allin Cottrell