obscure sampling problem fixed
by Allin Cottrell
We've recently fixed (in git and snapshots) a somewhat obscure
problem with the "smpl" command in gretl. I say "obscure" because
it's unlikely to affect most users, but here's the sort of thing
that could go wrong.
(1) You open or create a dataset.
(2) You impose a sample restriction that selects a non-contiguous
subset of observations.
(3) You then call a hansl function, to which one or more series are
passed as arguments, and in which "smpl" is used to modify the
sample range.
Now note that by gretl's design it should not be possible inside a
function to modify the sample to include observations that were
excluded at the level of the caller, as in step (2) above.
The following anomalies could arise in certain cases:
* Inside the function, when "smpl" is used more than once, on the
second or subsequent use the sample range might end up including
"excluded" observations.
* Inside the function, a valid use of "smpl" might fail with the
error message "No observations would be left" (i.e. gretl thinks the
sample range would be empty when that's not the case).
To our knowledge these points are now fixed, but if you have any
scripts that match the pattern above, please check that the results
look right. We want to make sure that we aren't missing any bad
cases.
Allin Cottrell
1 year, 4 months
- 1
- 0