| Re: Function names (was mean and sdev) [message #11041] |
Thu, 05 March 1998 00:00 |
wmc
Messages: 117
Registered: February 1995
|
Senior Member |
|
|
In article 2748ae80@p4.psych.uiuc.edu, "Kevin Spencer" <kspence1@uiuc.edu> writes:
> Brian Jackel <jackel@danlon.physics.uwo.ca> wrote in article
> <6dhu2f$eoi@ds2.acs.ucalgary.ca>...
>> Dr. G. Scott Lett wrote:
>>> The MOMENT function is obscure and slow when all you need is
>>> the mean and standard deviation, so we're adding the more obvious
>>> MEAN, STDDEV, VARIANCE and other statistics functions for IDL 5.1.
>> I'm not sure who the "we" above refers to, but I'd like to
>> make a general comment. IMHO IDL is already dangerously
>> cluttered with lots of special purpose functions that could
>> perhaps be more usefully bundled together.
> I've been very annoyed at RSI's inconsistent support for these
> simple statistical functions. There used to be a "stdev" procedure,
> then (if I recall correctly) one like it with a similar name. In IDL 5,
> they got rid of these functions and replaced them with "moment".
> Now according to Scott Lett, who I assume works for RSI, they
> are going to have another standard deviation procedure, plus one
> for computing the mean and another for the variance. Why did
> they get rid of stdev in the first place? And calculating the mean
> is trivial; just use total(x)/n_elements(x).
If they do introduce a function called stddev, instead of recalling stdev from
"obsolete", they will be very silly indeed (its still useable in IDL 5 BTW, just
not in the help).
There is disagreement between those who like things bundled together (the
"momentists") who are presumably happy to write av=total(x)/n_elements(x)
in order to avoid having a "mean" or "avg" function defined, and those who
don't really mind having piles of extra functions lying around in order to write
shorter and more comprehensible code (I'm in the latter camp).
To some extent this could be fixed by putting all the extra functions in a separate
directory whose name you could remove from the !PATH if you wanted to. Presumably,
if you don't want to see the obsolete functions you can remove "obsolete" from the
!PATH.
- William
---
William M Connolley | wmc@bas.ac.uk | http://www.nbs.ac.uk/public/icd/wmc/
Climate Modeller, British Antarctic Survey | Disclaimer: I speak for myself
|
|
|
|
| Re: Function names (was mean and sdev) [message #11050 is a reply to message #11041] |
Wed, 04 March 1998 00:00  |
Kevin Spencer
Messages: 1
Registered: March 1998
|
Junior Member |
|
|
Brian Jackel <jackel@danlon.physics.uwo.ca> wrote in article
<6dhu2f$eoi@ds2.acs.ucalgary.ca>...
> Dr. G. Scott Lett wrote:
>>
>> The MOMENT function is obscure and slow when all you need is
>> the mean and standard deviation, so we're adding the more obvious
>> MEAN, STDDEV, VARIANCE and other statistics functions for IDL 5.1.
>>
>> Best wishes,
>> Scott Lett
>
> I'm not sure who the "we" above refers to, but I'd like to
> make a general comment. IMHO IDL is already dangerously
> cluttered with lots of special purpose functions that could
> perhaps be more usefully bundled together. In the case of
> statistics it would be better to have a function like MOMENTS
> that could return any or all required moments. Even better
> would be something (perhaps called STATISTICS) that could
> provide a number of useful quantities. Although the use
> of square brackets should eliminate the confusion between
> functions and arrays, the proliferation of built-in functions
> (especially with common names) may cause problems if the
> name has been used for existing code.
>
> Does anyone else have an opinion on this subject?
I've been very annoyed at RSI's inconsistent support for these
simple statistical functions. There used to be a "stdev" procedure,
then (if I recall correctly) one like it with a similar name. In IDL 5,
they got rid of these functions and replaced them with "moment".
Now according to Scott Lett, who I assume works for RSI, they
are going to have another standard deviation procedure, plus one
for computing the mean and another for the variance. Why did
they get rid of stdev in the first place? And calculating the mean
is trivial; just use total(x)/n_elements(x). A separate function for
computing variance is likewise trivial and unnecessary, if there's
a function for computing the standard deviation.
Kevin
***************************************************
Kevin Spencer
Cognitive Psychophysiology Laboratory
Dept. of Psychology and Beckman Institute
University of Illinois at Urbana-Champaign
kspence1@uiuc.edu
***************************************************
|
|
|
|
| Re: Function names (was mean and sdev) [message #11053 is a reply to message #11050] |
Wed, 04 March 1998 00:00 |
Martin Schultz
Messages: 515
Registered: August 1997
|
Senior Member |
|
|
Brian Jackel wrote:
>
> Dr. G. Scott Lett wrote:
>>
>> The MOMENT function is obscure and slow when all you need is
>> the mean and standard deviation, so we're adding the more obvious
>> MEAN, STDDEV, VARIANCE and other statistics functions for IDL 5.1.
>>
>> Best wishes,
>> Scott Lett
>
> I'm not sure who the "we" above refers to, but I'd like to
> make a general comment. IMHO IDL is already dangerously
> cluttered with lots of special purpose functions that could
> perhaps be more usefully bundled together. In the case of
> statistics it would be better to have a function like MOMENTS
> that could return any or all required moments. Even better
> would be something (perhaps called STATISTICS) that could
> provide a number of useful quantities. Although the use
> of square brackets should eliminate the confusion between
> functions and arrays, the proliferation of built-in functions
> (especially with common names) may cause problems if the
> name has been used for existing code.
>
> Does anyone else have an opinion on this subject?
>
> --
> Brian Jackel
Oh yes, I do! In my opinion it gives much more readable code if you
can call a function named MEAN or STDDEV to get these values. On
the other hand, I see your point that there are far too many functions
in IDL that nobody can remember. But bundling always raises issues
of maintenance and completeness. My suggestion would be to think of
a number of "wrapper" routines (like your suggestion for statistics)
which can then take keyword parameters like /MEAN, /STDDEV, etc.
and which would call the appropriate internal IDL functions. But those
would not necessarily have to be developped by the IDL people ;-)
What I find much more bothering are inconsistencies within "families" of
functions, e.g. different ordering of parameters or naming of keywords
in say a linear fit and a polyfit function (of course, I cannot find a
suitable example right now, but I come across these things once a
while).
Just one additional word: the capability of finding the appropriate
IDL function depends strongly on the quality of the links in the online
help. After I had complained about a missing link to IDL once, they
replied and said they would be grateful for specific suggestions to
improve these links. At least there is some hierachical sorting in the
Navigate-Search Index menu - this could probably be improved (e.g.
"curve fitting" -- but why not "fitting" ?; how about an option to
search only the "topic" indices ?)
regards,
Martin.
------------------------------------------------------------ -------
Dr. Martin Schultz
Department for Earth&Planetary Sciences, Harvard University
186 Pierce Hall, 29 Oxford St., Cambridge, MA-02138, USA
phone: (617)-496-8318
fax : (617)-495-4551
e-mail: mgs@io.harvard.edu
IDL-homepage: http://www-as.harvard.edu/people/staff/mgs/idl/
------------------------------------------------------------ -------
|
|
|
|
| Re: Function names (was mean and sdev) [message #11065 is a reply to message #11050] |
Tue, 03 March 1998 00:00 |
J.D. Smith
Messages: 214
Registered: August 1996
|
Senior Member |
|
|
Brian Jackel wrote:
>
> Dr. G. Scott Lett wrote:
>>
>> The MOMENT function is obscure and slow when all you need is
>> the mean and standard deviation, so we're adding the more obvious
>> MEAN, STDDEV, VARIANCE and other statistics functions for IDL 5.1.
>>
>> Best wishes,
>> Scott Lett
>
> I'm not sure who the "we" above refers to, but I'd like to
> make a general comment. IMHO IDL is already dangerously
> cluttered with lots of special purpose functions that could
> perhaps be more usefully bundled together. In the case of
> statistics it would be better to have a function like MOMENTS
> that could return any or all required moments. Even better
> would be something (perhaps called STATISTICS) that could
> provide a number of useful quantities. Although the use
> of square brackets should eliminate the confusion between
> functions and arrays, the proliferation of built-in functions
> (especially with common names) may cause problems if the
> name has been used for existing code.
>
> Does anyone else have an opinion on this subject?
>
> --
> Brian Jackel
I agree wholeheartedly. It seems to me that making multipurpose
routines fast was the reason (among others) for introducing the function
arg_present. One can use it to make a version of MOMENT for instance,
that only calculates variance, skew and kurtosis if the user wants
them. I think in general a balance has to be struck between ease of use
and over-cluttering. In this case, since mean and stdev are likely the
most common calculations performed, they may merit their own functions,
but I do agree that restraint is in order (or a more robust shadowing
mechanism... but I won't hold my breath).
JD
--
J.D. Smith |*| WORK: (607) 255-5842
Cornell University Dept. of Astronomy |*| (607) 255-4083
206 Space Sciences Bldg. |*| FAX: (607) 255-5875
Ithaca, NY 14853 |*|
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
