Re: st: the memory footprint and execution speed of -mf_quadcross()-

[László Sándor <sandorl@gmail.com>]

Thanks again, Billy.

I was hoping to generate some discussion, esp. if we don't need to
speculate (like I could) but someone actually had a reference I did
not find or had some experience with the mentioned trade-off.

I did not expect anyone to present a cost-benefit analysis for my
case, of course, but I was (and am) lacking guidance on how much
slower can -mf_quadcross()- be. Not in my specific case, but in more
general scenarios.

But OK, let's be reasonable and say that if there are two separate
functions, the performance difference must be substantial, probably
even more for big data.

Stata does not do bad for my application, but yes, I want to improve
what I can. I don't think I brought up only freakish fringe cases.
E.g. to use st_views() without making copies could be important for
many applications, even with fewer observation and variables than what
I have. I still have no better answer.

Thanks!

Laszlo

On Thu, Sep 12, 2013 at 6:12 PM, William Buchanan
<william@williambuchanan.net> wrote:
> Hi Laszlo,
>
> No offense, but I think the reason that you don't elicit many responses isn't a fault of "usefulness of all [your] questions," but rather that you are just making a comment about how much Stata sucks for your applications.  You could probably generate some useful discussion (and quite possibly solve some issues that you're running into) if you framed the issues that you list in terms of questions with examples that other users can replicate and try to help solve.
>
> With regards to the two questions that you listed here (since the above is also meant to reference several of the other emails you sent after this), no one here could answer your first question.  Are you more interested in precision or speed?  If -mf_quadcross()- doesn't provide you with the speed that you want but gives you the precision that you need, then it wasn't a bad idea; conversely, if it provides a level of precision that you don't need but speed that is desirable then it also wouldn't be a bad idea.  So in an attempt to answer your second question, no one knows how bad because you've not clearly defined what bad or good are.
>
> HTH,
> Billy
>
> On Sep 12, 2013, at 2:52 PM, László Sándor <sandorl@gmail.com> wrote:
>
>> I checked some of Bill's resources on precision (e.g.
>> http://blog.stata.com/2012/04/02/the-penultimate-guide-to-precision/),
>> and -mf_quadcross()- is always mentioned as an option, possibly
>> overkill, but without much guidance on what the costs of its use might
>> be in terms of memory or execution speed.
>>
>> If my code uses the function many times on humungous st_views, was it
>> a bad idea to use quadcross()? Probably it was — but how bad?
>>
>> Thanks,
>>
>> Laszlo
>>
>> PS: I must wonder about the usefulness of all my questions if they
>> attract few comments. Thanks for bearing with me anyway.
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>> *   http://www.ats.ucla.edu/stat/stata/
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/