[R] Saving fits (glm, nls) without data

[David Winsemius]

On Sep 7, 2010, at 2:53 PM, Johann Hibschman wrote:

> David Winsemius <dwinsemius at comcast.net> writes:
>
>> On Sep 7, 2010, at 11:02 AM, Johann Hibschman wrote:
>>> Even so, I would prefer to only save the coefficients
>>
>> Have you read through the Value section of glm's help page?
>>
>> ...and
>>
>> ?coef
>
> I have; it's easy to get the coefficients. The part I'm struggling  
> with
> is reconstituting an operational glm object given the coefficient  
> vector
> and the formula.
>
> Really, I was hoping that someone had already done this work, so I  
> could
> stop trying re-create the right kind of term object, while making sure
> it doesn't hold on to a pointer to an environment with a lot of data  
> in
> it, etc., etc..

I was assuming you could take all the code work that was already  
tested and trim out the non essential code and arguments let it work  
on a new class.
>
>> The predict.glm function is visible so you can just type its name to
>> see the code. It appears that the section of the code that does the
>> work is fairly short. This is my nomination for what happens in most
>> cases:
>
>> if (!se.fit) {# not generally invoked with se.fit=TRUE
         if (missing(newdata)) {  # forgot this clause in first post
>>        }
>>        else {
>>            pred <- predict.lm(object, newdata, se.fit, scale = 1,
>>                type = ifelse(type == "link", "response", type),
>>                terms = terms, na.action = na.action)
>>            switch(type, response = {
>>                pred <- family(object)$linkinv(pred)
>>            }, link = , terms = )
>>        }
>
> I agree.  That reduces the problem to confecting a working lm object,
> given a formula and coefficients.  Unfortunately, I haven't yet  
> figured
> out how to do that.

And I was thinking one would start with the glm object and just set  
the unnecessary leaves of the list to NULL.

>
>> So maybe you should write a predict function that would work on a
>> reduced glm object that has a class name of your choosing.
>
> I'm trying to get this to work, but I haven't figured out yet how to
> generate the X matrix properly from the formula and the coefficients.
> I'm sure I can eventually get it, but it's annoying.

I don't think you need to do anything other than construct a proper  
newdata argument and feed it and your stripped down object to a  
modified predict.sml_glm function. And it could very well be that all  
you need to do is rename the predict.glm code and give it the proper  
arguments. I do not see a need to recreate an X matrix for the  
newdata, since the code to do that is already in predict.lm(). I  
suppose I could be wrong, since I have not done it myself. I just got  
my employer to buy more memory.

>
> The whole model whereby fit objects keep around their data so you  
> don't
> have to provide it on a few calls just seems like a mistake.
>
> -Johann
>

David Winsemius, MD
West Hartford, CT