subscripting a terms object

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

subscripting a terms object

R devel mailing list
Someone sent me a bug report for survival2.44.1-1 that involves a model with both cluster
and offset.  It turns out to be a 3 part issue with [.terms and my own untangle.specials
routine.   I've spent an evening sorting out the details.


   1. The delete.response() function doesn't remove the response from the dataClasses
attribute, which leads to a later failure in [.terms for no-response models.  Is there a
reason for this, or can I make my patch include this oversight as well?

  2. [.terms messes up predvars and dataClasses if the model has an offset term in it. 
(In select cases 1 and 2 can cancel out and give the correct dataClasses attribute.)

  3. The survival::untangle.specials routine assumed that you can use the same
subscripting for the terms of a model and the term() object itself, which turns out to be
almost always true, but only almost.

   The failure turns out to have probably been there since the Splus days, which tells one
just how often such a model is used. (One of two edge case bugs sent to me in the first
days after I pushed it to CRAN: a new release seems to attact them.)   I'm willing to put
together a patch, but given the rarity of these would folks prefer to wait until after the
April release?   I'm fine with that.  I need the answer to 1 though.

Terry T.


        [[alternative HTML version deleted]]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Reply | Threaded
Open this post in threaded view
|

Re: subscripting a terms object

Martin Maechler
Dear Terry,

>>>>> Therneau, Terry M , Ph D via R-devel
>>>>>     on Thu, 4 Apr 2019 22:48:49 -0400 writes:

    > Someone sent me a bug report for survival2.44.1-1 that involves a model with both cluster
    > and offset.  It turns out to be a 3 part issue with [.terms and my own untangle.specials
    > routine.   I've spent an evening sorting out the details.



    >   1. The delete.response() function doesn't remove the response from the dataClasses
    > attribute, which leads to a later failure in [.terms for no-response models.  Is there a
    > reason for this, or can I make my patch include this oversight as well?

    >  2. [.terms messes up predvars and dataClasses if the model has an offset term in it. 
    > (In select cases 1 and 2 can cancel out and give the correct dataClasses attribute.)

The above two seem interesting and relevant to R itself.
As we've recently just fixed a buglet in  reformulate() --
probably unrelated to your problem --  I'd really be interested to see a
repr.ex. (reproducible example) for the above two statements.

... and if you want also a proposal on how to address the
problem in  delete.response()  and/or  `[.terms`()

Martin

    >  3. The survival::untangle.specials routine assumed that you can use the same
    > subscripting for the terms of a model and the term() object itself, which turns out to be
    > almost always true, but only almost.

    >   The failure turns out to have probably been there since the Splus days, which tells one
    > just how often such a model is used. (One of two edge case bugs sent to me in the first
    > days after I pushed it to CRAN: a new release seems to attact them.)   I'm willing to put
    > together a patch, but given the rarity of these would folks prefer to wait until after the
    > April release?   I'm fine with that.  I need the answer to 1 though.

    > Terry T.

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel