I recently became disillusioned with gnuplot and overall suggest prodding and 
helping the PLplot developers to improve their product.  The breaking point for 
me was plotting ~500k sample time series and watching Pharo+gnuplot struggle 
with marshaling data into and out of text or worse, fighting with barely 
documented methods for moving binary data into gnuplot.  Remember my questions 
about a Linux analog to Windows device contexts (one of the few things MS got 
right)?  Ground zero for this.  

My PLplot interface will likely appear "soon."  It needs to be cleaned up with 
1.1.1 in mind, as does (much more so) code that I will make available for 
AccesIO USB A/D hardware.  

For numerical calculations, I agree that well-worn code (which can be hard to 
find) is of great value.  GSL is pretty good.  One thing that irritates me no 
end is that they use their vectors and matrices just enough to be a major pain 
but not at all consistently.  They also need to be scolded for making .so's 
(GSL and BLAS) that can't be loaded separately - they know about it - don't 
hold your breath.  Callbacks are sorely needed on our end[*].  When I have 
something worthy of release, I'll let you know; no promises about Windows 
support, though I won't deliberatly sabotage it - I simply don't care (and it 
feels GOOD<g>).

R is both friend and foe.  In general, I prepare data using Smalltalk (so I 
widen or lengthen data frames in Pharo rather than trying to figure out R 
syntax for such things).  Rejecting garbage, to a point, happens in R. I 
generally prefer to let R arrive at statistical measures (p-values, correlation 
coeffs, etc.) that are going to leave the nest.  My late Friday PM submission 
(now we wait...) contains numbers generated with various libraries, massaged in 
R, graphs from R and PLplot and nothing from gnuplot (which I was using just  a 
couple of months ago).  One scary thing: PLplot is only now growing a 
formalized way to create legends, which gnuplot does very well.  But if you 
deal with large numbers of points, gnuplot can become slow and pedantic.  The 
usual excuses such as "if you have that much data, you can't see them 
anyway..." don't really cut it.  There is something to be said for plotting 
everything and letting what Tufte would call ink density tell part of the 
story, especially with time series where peaks/outliers are interesting 
(downsampling will generally distort the message).

[*] Sig, yes, I know... :)  With the release of 1.1.1 and completion (for now) 
of my grant proposal, I need to start looking at undoing any damage that I have 
caused in the name of avoiding underscores.  Some things are arguably nicer for 
it, but most suffered mightily.  I am thinking exclusively of exernal 
interfacing; some of the stuff was named in a way that expects underscore to be 
present, and removing them makes a hell of a mess.  Time to fix that (thanks to 
all how have made it possible!!!), get daily operations on 1.1.1 and then to 
begin playing with NativeBoost.  I see there are pre-built Cog vms.  Is there a 
way to either build Cog+NB or to transfer the NB plugin built elsewhere to Cog? 
 Sounds like a good problem to have.

Bill



-----Original Message-----
From: [email protected] 
[mailto:[email protected]] On Behalf Of 
[email protected]
Sent: Saturday, October 02, 2010 7:29 PM
To: [email protected]
Subject: Re: [Pharo-project] Statistic tools in Pharo

I've been lurking on this thread because I also need a lot of statistics in my 
work, but as already wrote in another thread on far but related subject: I 
think we are better of if we get the FFI or Alien or its successor to work in 
such a way that makes it easy to interface with existing, thoroughly tested and 
continuously developed functionality for these specific areas.

I presently work a lot with R http://www.r-project.org/ and I remember an 
embryonic interface to it was done in Dolphin Smalltalk (which is Windows 
only), besides seem a former associate testing it a little, I went not any 
further on exploring it.

BTW, going to wishful mode I _think_ having connection to packages of this 
calibre (another one I'll like to have easy access is GNUPlot) would be even 
better if we could get the image inside a morphic object which would work as a 
graphics device for these programs. . .

my .0199999. . .

[]s

 



Em 02/10/2010 11:24, Alexandre Bergel < [email protected] > escreveu:
Ok. I am personally more interested in regression models

Alexandre


On 2 Oct 2010, at 07:15, Cédrick Béler wrote:

> 
> Le 29 sept. 2010 à 15:39, Alexandre Bergel a écrit :
> 
>>>> Thanks Jimmie. 
>>>> About half of the tests are red. This library is about linear Algebra. Not 
>>>> really about statistics. 
>>>> But thanks for the link!
>>>> 
>>>> Alexandre
>>> 
>>> If it's the same as what you can find in VW, in addition to linear algebra, 
>>> it includes numerical methods for quite a few probability distributions as 
>>> well though.
>> 
>> Ok, thanks for the input!
> 
> 
> FWIW, I plan to implement Bayesian Network in Smalltalk... Nevertheless, it 
> will be more a research project...
> 
> Until now, I haven't found any project nor code in that direction... except 
> maybe SBN (Simple Bayesian Network) in ruby...
> 
> Cheers
> 
> 
> 
> 
> 
>> 
>> Alexandre
>> --
>> _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:
>> Alexandre Bergel  http://www.bergel.eu 
>> ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
>> 
>> 
>> 
>> 
>> 
>> 
>> _______________________________________________
>> Pharo-project mailing list
>> [email protected]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
> 
> 
> _______________________________________________
> Pharo-project mailing list
> [email protected]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

--
_,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:
Alexandre Bergel  http://www.bergel.eu
^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.






_______________________________________________
Pharo-project mailing list
[email protected]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project


_______________________________________________
Pharo-project mailing list
[email protected]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
_______________________________________________
Pharo-project mailing list
[email protected]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

Reply via email to