Re: CLEAR TRANSFORMATIONS ?

[Ben Pfaff]

[Sorry, I fell off the net for a few days when my wife and I were
moving from one house to another and our Internet service hadn't
yet been transferred.  I will probably be sporadic for another
couple of days.]

John Darrington <address@hidden> writes:

> On Mon, Jul 02, 2007 at 07:41:01AM -0700, Ben Pfaff wrote:
>      John Darrington <address@hidden> writes:
>      
>      > On Sat, Jun 30, 2007 at 07:55:27AM -0700, Ben Pfaff wrote:
>      >      There are a few subtleties I think.  First, I don't think we save
>      >      a copy of the dictionary that corresponds to the active file.
>      >      We'd need to do so, and then restore it for CLEAR
>      >      TRANSFORMATIONS.
>      >
>      > Spss doesn't do this.  So neither should we (at least not by default).
>      
>      I think that SPSS does do this.  The SPSS 13, 14, and 15 manuals
>      all have an identical example that shows a data file being read,
>      some transformations being done on it, and then running CLEAR
>      TRANSFORMATIONS.  The text then states that the variables created
>      by the transformations no longer exist, because they were created
>      by transformations that have been discarded.
>
> You're right!  The manuals do have this example.  I tried running that
> example on v15 and it does exactly what the manual says it won't do!!
> It DOES NOT remove the INDEXQ variable! (see attached output file).

I have one question first: in what way did you make SPSS run that
syntax?  There is a distinction between "interactive" and "batch"
environments made in the manual.  I see the documented behavior
on a very old version of SPSS here: the variable is dropped in
interactive mode, CLEAR TRANSFORMATIONS just yields a warning in
batch mode.

> So I suppose the question for us is, do we follow the manual or our
> observations ?

I can't see how this is anything but a bug in SPSS, so I would
rather follow the documentation.

>      The transformation that I'm talking about does persist after the
>      command terminates.  Take a look at cmd_descriptives and where
>      the call to setup_z_trns is located.
>
> Hmm.  I can see that this might cause problems for the GUI anyway.
> When the descriptives command returns, the values of the z-scores will
> not have been populated.
>
> I can see that for optimisation reasons it's desireable not to execute
> this transformation when running from a script, but the GUI needs to
> run it, when it redraws the data sheet.  Otherwise the user will get
> confused.   So maybe these "hidden" transformations should be
> specially marked as such (perhaps in a seperate transformation chain)?

Yes, I agree that they should be marked.  A separate
transformation chain is a reasonable implementation choice, I
think.

I am not sure that the GUI really needs this, assuming that it
runs something like proc_execute() before updating the data
sheet.  If it does that, then any "hidden" transformations will
be executed on the spot; if it doesn't, it seems like even an
ordinary transformation like COMPUTE wouldn't update the data
sheet whether it created a new variable or not.  So my guess is
that it is important only for CLEAR TRANSFORMATIONS.
-- 
Ben Pfaff 
http://benpfaff.org