jnumerical-general — A generic mailing list

[Jnumerical-general] planning to fork project

[Frank W. <fwi...@gm...>]

Hi all,

I received a patch on the Jython list that updates this project to
work with Jython 2.2.  I'd also like to see this project work with the
coming 2.5.  I am thinking of pulling it over to a different "forge"
that supports DVCS (probably kenai because I like mercurial).  Since
this project hasn't seen an update for many years, I doubt I'm
stepping on anyone's toes -- but I don't want to be rude.  Of course
any of the committers on this project would be welcome on the new
site, or if there are strong feelings about staying here, I'd be happy
to talk about joining this project here instead.

-Frank

RE: [Jnumerical-general] jnumeric future

[zubin <bin...@be...>]

Jingzhao, any ideas on how we can resurrect this project?

-----Original Message-----
From: jnu...@li...
[mailto:jnu...@li...]On Behalf Of
Jingzhao Ou
Sent: Sunday, April 18, 2004 7:11 PM
To: zubin; jnu...@li...
Subject: Re: [Jnumerical-general] jnumeric future


Hi,

I think that jNumeric is not active for quite some time. I really would like
to
see that it can be active again as I also like this project very much.

I am willing to help out if possible.

Best regards,
Jingzhao

==========================================
Jingzhao Ou
Ph.D. Candidate, Department of Electrical Engineering
University of Southern California
Los Angeles, CA, USA
http://www-scf.usc.edu/~ouj

--- zubin <bin...@be...> wrote:
> hello, i am a corporate user of jython, think the jnumeric project is
great.
> Have $ to support.  I just need to know if it is being supported, is this
> still an active project?  Just trying to get some feeling on the future of
> this.
>




-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
Jnumerical-general mailing list
Jnu...@li...
https://lists.sourceforge.net/lists/listinfo/jnumerical-general

Re: [Jnumerical-general] jnumeric future

[Jingzhao Ou <ja...@ya...>]

Hi,

I think that jNumeric is not active for quite some time. I really would like to
see that it can be active again as I also like this project very much. 

I am willing to help out if possible. 

Best regards,
Jingzhao

==========================================
Jingzhao Ou
Ph.D. Candidate, Department of Electrical Engineering
University of Southern California
Los Angeles, CA, USA
http://www-scf.usc.edu/~ouj

--- zubin <bin...@be...> wrote:
> hello, i am a corporate user of jython, think the jnumeric project is great.
> Have $ to support.  I just need to know if it is being supported, is this
> still an active project?  Just trying to get some feeling on the future of
> this.
> 

[Jnumerical-general] jnumeric future

[zubin <bin...@be...>]

hello, i am a corporate user of jython, think the jnumeric project is great.
Have $ to support.  I just need to know if it is being supported, is this
still an active project?  Just trying to get some feeling on the future of
this.

[Jnumerical-general] A forwarded message

[Jingzhao Ou <jin...@sb...>]

Great. Since you plan to do some hacking, allow me to comment as to where the
community should take JNumeric.

The next step in JNumeric is to actually work backward. Initially, JNumeric was
implemented as all-Java project. This proves to be counterproductive in
practice. Indeed, if you look at Python Numeric, most of the implementation is
done in Python with a few C calls. That's how JNumeric should work: mostly in
Python with very few Java calls.

Ideally, we should share nearly the same Python code with Python Numeric. This
way, we would have far fewer lines of Java code to maintain and we would
immediately get most missing features from 
Python Numeric.

In fact, if you study the code, most of the serious things we are missing are
the linear algebra routines. However, there are many good librairies in java to
do linear algebra, so it isn't hard to implement these missing features.

(Alas, I don't do any linear algebra these days, so I'm not a 
good candidate to implement those.)

So, basically, here's what I suggest: if you need to extend or fix JNumeric,
please try to do as much as possible in Python. It was a mistake to do too much
in Java.

-D


>Dear Dr. Lemire,
>
>Thanks a lot for your prompt response. Let me make myself more clear.
>
>First, I am now using the CVS version of Jython on my Windows XP 
>machine.
When I type "jython" on my Cygwin, it displays the following:
>
>$ jython
>Jython 2.1+ on java1.4.1_02 (JIT: null)
>Type "copyright", "credits" or "license" for more information.
>>>>
>
>After I installed jNumeric, I tested it using the following and every
thing is fine.
>
>>>> import Numeric
>>>> Numeric.pi
>3.141592653589793
>>>> a = Numeric.ones(23)
>>>> a
>array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
>1, 1])
>>>> Numeric.sum(a)
>23
>
>When I look at the source code of jNumeric, I can see the following
>
>public class JNumeric implements InitModule {
>
>	public void initModule(PyObject dict) {
>
>		// import modules
>		dict.__setitem__("umath", PyJavaClass.lookup(Umath.class));
>		dict.__setitem__("FFT", PyJavaClass.lookup(JN_FFT.class));
>
>......
>
>However, if you go to http://www.jython.org/docs/javadoc/, you can see 
>the
following
>
>===================================================================
>public interface InitModule
>An deprecated interface that can be used if a java class want control 
>over
the class dict initialization. 
>
>This class is deprecated. See ClassDictInit for a replacement.
>===================================================================
>
>When I looked at the source code of the CVS version of Jython, I can 
>see
that they are actually using ClassDictInit already.
>
>===================================================================
>public class PyList extends PySequence implements ClassDictInit {
>     public static void classDictInit(PyObject dict) {
>        PySequence.classDictInit(dict);
>===================================================================
>
>So, I am not sure whether jNumeric should reflect such changes or not.
Would you please give me some guidance on this? Thanks!
>
>I have been using Python for quite some time. Recently, I become very
interested in Jython and jNumeric. Especially, I hope that I can do some thing
for jNumeric. 
>
>I am a Ph.D. student at University of Southern California, Los Angeles,
USA. I am now working on some beamforming algorithms used in wireless systems
and their implementation using FPGAs. I hope that I can enhance jNumeric a
little bit with regard to what I am doing right now. 
>
>I am pouring over the source code of both Jython and jNumeric. Hopeful
that I make my contribution to jNumeric soon. 
>
>Have a nice day!
>
>Best regards,
>Jingzhao
>
>----- Original Message -----
>From: Daniel Lemire <le...@on...>
>Date: Monday, June 9, 2003 11:58 am
>Subject: Re: About the status of jNumeric
>
>> Not sure what you mean. I've been using Jython 2.1 with JNumeric for 
>> a couple of years now with no problem. Just went to the Jython web 
>> site, and that's still the current version.
>> 
>> If you found a deprecated feature and wish to help us keep JNumeric 
>> current, I'd gladly create a developer's account for you.
>> 
>> 
>> --
>> Daniel Lemire, Ph.D.
>> Research Officer, National Research Council of Canada
>> Adjunct Professor, University of New Brunswick
>> http://www.ondelette.com/acadia/
>
>
>

[Jnumerical-general] Is jNumeric alive?

[Jingzhao Ou <ja...@ya...>]

Hi,

I wonder if jNumeric is still alive? Thanks!

Best regards,
Jingzhao

[Jnumerical-general] How to get access to the CVS jNumeric?

[Jingzhao Ou <jin...@sb...>]

Hi, 

I am very interested in jNumeric and would like to add some digital signal
processing computation abilities to it. However, when I downloaed the CVS
jNumeric according the instructions on the website, it seems to me that it is
not the stuff I want. Can any one tell me how to download the correct files?

BTW, is there any one still developing jNumeric? 

Thanks!

Best regards,
Jingzhao

[Jnumerical-general] Linear Algebra

[Daniel L. <le...@on...>]

Thank you Drew for your message. I don't think the fact that you use Colt is 
something to feel sorry about. 

The whole intent of Jython is not to make things like Colt obselete. The 
intent would be to build Python wrappers around them.

There are other libraries in Java competing with Colt however, one by IBM and 
JSci.

-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
Adjunct Professor, University of New Brunswick
http://www.ondelette.com/acadia/

[Jnumerical-general] Re: Linear Algebra

[<so...@st...>]

Hi Drew,
I also use Colt for statistical calculations (drawing numbers from 
statstical populations). However, I use Jame for Linear Algebra.
From what I remember the Linear Algebra routines in Colt were a copy of 
Jama.

I resently followed a discussion in this mailing list where Brian Zimmer 
was working on implementing Linear Algebra in Jnumeric (using BLAS 
routines and f2j). I look forward to see those results.

regards,
Even Solbraa

[Jnumerical-general] Re: Jnumerical-general digest, Vol 1 #37 - 1 msg

[Frank G. <fgi...@hm...>]

At 03:20 PM 1/21/2003, you wrote:
>Anyway, in the
>meantime, from recommendations of two friends who do pretty serious
>scientific computing, I started using COLT for my calculations. It really
>seems to do a lot of the math quite well and very quickly.

That sounds interesting, Drew. Is this what you're talking about?

http://hoschek.home.cern.ch/hoschek/colt/

I checked out the Introduction - sounds pretty awsome! Thanks for the 
recommendation!

-Frank


PhD, Computational Biologist,
Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115, USA.
Tel: 617-432-3555       Fax: 
617-432-3557       http://llama.med.harvard.edu/~fgibbons

[Jnumerical-general] Linear Algebra

[Drew B. <dba...@ri...>]

Hi all,
    I was originally very interested in Jnumeric until I discovered that
there was essentially no work in the Linear Algebra area. Frankly, this area
particularly worries me--- I believe it is very hard to do numerical linear
algebra well. Numeric isn't worth much to me without things like svd(), but
I understand it has more broad appeal to many others. Anyway, in the
meantime, from recommendations of two friends who do pretty serious
scientific computing, I started using COLT for my calculations. It really
seems to do a lot of the math quite well and very quickly. It is composed of
quite a few parts, and as Daniel rightly pointed out to me, some of the less
important parts may have liscensing issues. Nevertheless it may very well
make a strong base for a Linear Algebra component to Jnumeric.
    Sorry I haven't been of more help, but I thought the pointer to COLT
might be valuable. Best of luck,

    yours,
    --drew

[Jnumerical-general] Re: Re: Porting issues and Jython 2.2

[Daniel L. <le...@on...>]

> True division means '1/2' will return '0.5' not '0'.  Jython supports this
> in CVS.

Ok, and 1/2=0 is a "false" division then. ;-)

> There's only four maintainers.  Two of us don't have much to do with the
> core language changes such as new style classes and generators so there's
> not alot of development time available to such big changes.  It's really
> too bad so few resources are available.

Right. Still, Jython seems to fare a lot better than the lack of releases 
suggest. Cool.

RE: [Jnumerical-general] Re: Porting issues and Jython 2.2

[brian z. <bz...@zi...>]

> According to the changes.txt of Numeric 21.0, Numeric now implements true
> division operatons relying on Python 2.2 whatever this means. I
> have no clue,
> but it seems like Numeric uses the new features of Python 2.2.
>

True division means '1/2' will return '0.5' not '0'.  Jython supports this
in CVS.

> I don't use CVS Jython so I'll take your word for it. Does it
> mean that Jython
> now implements iterators? I seem to recall that it was the big change.
>

Jython in CVS does support iterators.  The core objects have been updated to
make use of the new API.

> How is Jython going along? Jython seems frozen as far as I can
> tell. Jython as
> it is, is good enough for my purposes, but if we are going to
> rely on it more
> heavily to keep up with Python, then we should know that it is
> doing well.
> The mailing list seems active, but I don't see many releases.
>

There's only four maintainers.  Two of us don't have much to do with the
core language changes such as new style classes and generators so there's
not alot of development time available to such big changes.  It's really too
bad so few resources are available.

brian

[Jnumerical-general] Numeric vs numarray

[Frank G. <fgi...@hm...>]

At 03:21 PM 1/7/2003, you wrote:
> > Does anyone know about the upcoming migration from Numeric to Numarray?
> > What affect will this have JNumeric?
>
>I have no clue. I have never used Numarray.


I asked Paul Dubois about this a few months back, because I was confused: 
they're not fixing bugs in Numeric anymore, because Numarray will supercede 
it, but Numarray is currently only at release 0.4 (or thereabouts), and 
they're still arguing over what appear to me to be some very fundamental 
issues of syntax (to implement '//', e.g.). Dubois's answer was that 
Numarray is further along than it appears, but there's resistance to 
increasing the release number.... So, I guess, you should take claims of 
its imminent release with a pinch of salt.

-Frank


PhD, Computational Biologist,
Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115, USA.
Tel: 617-432-3555       Fax: 
617-432-3557       http://llama.med.harvard.edu/~fgibbons

[Jnumerical-general] Re: Porting issues and Jython 2.2

[Daniel L. <le...@on...>]

Yes, please do not take my comments for a criticism of your approach. It seems 
well founded, a h*ll of a lot better than what we have right now it terms of 
maintenance. Beside, none of us on this list are responsible for the current 
design choice... so nobody will fight you if you improve the design!!!

>  Versions after 20.0 require Python 2.0 or later. To take advantage of
>  the "rich comparisons" (i.e., to be able to compare arrays and get
>  back a boolean result) you need Python 2.1.

According to the changes.txt of Numeric 21.0, Numeric now implements true 
division operatons relying on Python 2.2 whatever this means. I have no clue, 
but it seems like Numeric uses the new features of Python 2.2.

> I agree completely with your concerns about Numeric eventually requiring >
> Python 2.2 but the only *big* difference between CVS jython and Python 2.2
> is new style classes AFAIK.

I don't use CVS Jython so I'll take your word for it. Does it mean that Jython 
now implements iterators? I seem to recall that it was the big change. 

How is Jython going along? Jython seems frozen as far as I can tell. Jython as 
it is, is good enough for my purposes, but if we are going to rely on it more 
heavily to keep up with Python, then we should know that it is doing well. 
The mailing list seems active, but I don't see many releases.

> Does anyone know about the upcoming migration from Numeric to Numarray?
> What affect will this have JNumeric?

I have no clue. I have never used Numarray.

RE: [Jnumerical-general] Porting issues and Jython 2.2

[brian z. <bz...@zi...>]

All,

From the README included with Numeric (I don't have Python 2.0 to confirm):

 Versions after 20.0 require Python 2.0 or later. To take advantage of
 the "rich comparisons" (i.e., to be able to compare arrays and get
 back a boolean result) you need Python 2.1.

I agree completely with your concerns about Numeric eventually requiring >
Python 2.2 but the only *big* difference between CVS jython and Python 2.2
is new style classes AFAIK.  I don't think Numeric makes use of this new
feature though it certainly could.

In any case I'm going to look through the source more thoroughly to see how
much of an effort it would require to refactor parts of JNumeric to make use
of Numeric's Python interface.  Most of multiarray is exposed in
PyMultiarray as static methods and some of the other required C modules are
implemented in part in Java as well.  I'd really like to see LinearAlgebra
implemented as a lapack_lite module rather than re-write LinearAlgebra.py.
We'll see.

Does anyone know about the upcoming migration from Numeric to Numarray?
What affect will this have JNumeric?

thanks,

brian

> -----Original Message-----
> From: jnu...@li...
> [mailto:jnu...@li...]On Behalf Of
> Daniel Lemire
> Sent: Tuesday, January 07, 2003 9:16 AM
> To: jnu...@li...;
> jnu...@li...
> Subject: [Jnumerical-general] Porting issues and Jython 2.2
>
>
> Good day,
>
> It seems to me that your way would be much better.
>
> One issue I see though is that jython is basically frozen since
> January 2001,
> 2 years ago at version 2.1, whereas Python is constantly evolving
> (they are
> at version 2.3) so that it might not be so simple. Anybody knows
> when Jython
> 2.2 will come out?
>
> The latest Numeric relies on 2.2 as far as I know. It might still
> run in 2.1,
> I don't know... but I'm not sure. For sure, it is only a matter of time
> before they require 2.2.
>
>
>
>
> > On that topic, why was it chosen to reimplement Numeric.py in Java
> > rather than just re-use the existing Python-based modules included in
> > Numeric and just making sure that the modules originally written in C
> > (multiarray for instance) are properly exposed through Java so Jython
> > can access them.  This does require a Jython user to download and
> > install Numeric but it would make for less work on JNumeric's side as
> > well as forcing API consistency (so long as all methods are implemented
> > but this problem exists even if re-written in Java) with Numeric.  I
> > chose this approach when porting mx.DateTime to Jython [4] and it made
> > for a lot less work.
>
>
> -------------------------------------------------------
> This SF.NET email is sponsored by:
> SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
> http://www.vasoftware.com
> _______________________________________________
> Jnumerical-general mailing list
> Jnu...@li...
> https://lists.sourceforge.net/lists/listinfo/jnumerical-general

[Jnumerical-general] Porting issues and Jython 2.2

[Daniel L. <le...@on...>]

Good day,

It seems to me that your way would be much better. 

One issue I see though is that jython is basically frozen since January 2001, 
2 years ago at version 2.1, whereas Python is constantly evolving (they are 
at version 2.3) so that it might not be so simple. Anybody knows when Jython 
2.2 will come out?

The latest Numeric relies on 2.2 as far as I know. It might still run in 2.1, 
I don't know... but I'm not sure. For sure, it is only a matter of time 
before they require 2.2.




> On that topic, why was it chosen to reimplement Numeric.py in Java
> rather than just re-use the existing Python-based modules included in
> Numeric and just making sure that the modules originally written in C
> (multiarray for instance) are properly exposed through Java so Jython
> can access them.  This does require a Jython user to download and
> install Numeric but it would make for less work on JNumeric's side as
> well as forcing API consistency (so long as all methods are implemented
> but this problem exists even if re-written in Java) with Numeric.  I
> chose this approach when porting mx.DateTime to Jython [4] and it made
> for a lot less work.

[Jnumerical-general] implementing Linear Algebra

[Frank G. <fgi...@hm...>]

Hi Brian,


>I know Numeric uses LAPACK but I don't have much experience with the
>Java port [3] of LAPACK (currently in alpha) but it would be consistent
>with Numeric.  Any suggestions/comments?  I'm most interested in using
>the LAPACK port for consistency and minimizing development time unless
>for some reason it just doesn't work well with Jython (either
>programmatically or because of licensing).  If the required lapack_lite
>methods needed by LinearAlgebra.py are exposed we can even reuse
>Numeric's version.

>On that topic, why was it chosen to reimplement Numeric.py in Java
>rather than just re-use the existing Python-based modules included in
>Numeric and just making sure that the modules originally written in C
>(multiarray for instance) are properly exposed through Java so Jython
>can access them.

The folks at SciPy have imported pretty much the entire pantheon of 
numerical code in FORTRAN and C from the 1970's onward (ODEPACK, FITPACK, 
LAPACK, etc.), for use with Python (but not Jython). In view of what you're 
saying in the last paragraph, perhaps some kind of collaboration with them 
might be a smoother way to go?

As for the design decisions, I'm a late-comer to Numeric and Jython (early 
2002), at which point it was a done deal. The way you describe sounds 
cleaner and more maintainable.

-Frank



PhD, Computational Biologist,
Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115, USA.
Tel: 617-432-3555       Fax: 
617-432-3557       http://llama.med.harvard.edu/~fgibbons

[Jnumerical-general] implementing LinearAlgebra

[brian z. <bz...@zi...>]

Hello,

I'm interested in implementing the LinearAlgebra module for JNumeric.  I
recently needed the solve_linear_equations function for my own purposes
so I made a simple wrapper [1] around JAMA [2].  While this worked it
requires an external dependency and I wanted to know the group's
preference on external libraries.  If one is used does anyone have a
preference?

I know Numeric uses LAPACK but I don't have much experience with the
Java port [3] of LAPACK (currently in alpha) but it would be consistent
with Numeric.  Any suggestions/comments?  I'm most interested in using
the LAPACK port for consistency and minimizing development time unless
for some reason it just doesn't work well with Jython (either
programmatically or because of licensing).  If the required lapack_lite
methods needed by LinearAlgebra.py are exposed we can even reuse
Numeric's version.

On that topic, why was it chosen to reimplement Numeric.py in Java
rather than just re-use the existing Python-based modules included in
Numeric and just making sure that the modules originally written in C
(multiarray for instance) are properly exposed through Java so Jython
can access them.  This does require a Jython user to download and
install Numeric but it would make for less work on JNumeric's side as
well as forcing API consistency (so long as all methods are implemented
but this problem exists even if re-written in Java) with Numeric.  I
chose this approach when porting mx.DateTime to Jython [4] and it made
for a lot less work.

In addition I'd like to check in an Ant build.xml that parallels the
Makefile in src if no one objects.

thanks,

brian

[1] http://tinyurl.com/3t8o
[2] http://math.nist.gov/javanumerics/jama/
[3] http://sourceforge.net/projects/f2j
[4] http://sourceforge.net/zxpy/

[Jnumerical-general] Re: JNumeric LinearAlgebra

[Daniel L. <le...@on...>]

I had never noticed that bit, but given that I never had to change or even 
read this part of the code base, it is not surprising. It is not the most 
exciting part. We would have to check with Timothy Hochberg, but frankly, one 
can rather easily generate this by hand using a clever mix of copy/paste, 
search/replace. I guess you can do it in perl too... a matter of taste I 
suppose. I know that gnu.trove is entirely generated using perl scripts. 
That's one way to go.

So, if you are interested, please try to get in touch with Timothy.

Alternatively, ask Frank who was the last one to play with that file, he might 
know something about it.

Cheers!



> As for "autogenerated", looking through PyMultiarray I noticed that most of
> the magic methods are defined between two comments "// Begin generated code
> (genOps)." and "// End generated code (genOps)."  I assumed these to be
> autogenerated.
>
> > -----Original Message-----
> > From: Daniel Lemire [mailto:le...@on...]
> > Sent: Thursday, January 02, 2003 2:37 PM
> > To: brian zimmer
> > Cc: jnu...@li...
> > Subject: Re: JNumeric LinearAlgebra
> >
> >
> > I'm not sure what you mean by "autogenerated". The stuff I have
> > here is hand
> > coded... there is no perl engine if that's what you are referring
> > to... not
> > that I know. But other people have been active on the CVS tree...
> > Oh... maybe
> > you mean the regression testing? Anyhow, there is nothing I
> > worked on that
> > was "autogenerated", we'll have to track who submitted this...
> >
> > The best thing to do at this point is join the mailing list if
> > you haven't
> > done so. There is a jnumeric mailing list on sourceforge. It has not been
> > very active, but it is alive. I stronly suggest that you forward
> > (as c.c.)
> > any question or answer to that list so that future generations
> > can benefit
> > from our collective wisdom.
> >
> > I was very active last year, then Frank Gibbons took over and he has been
> > hacking at the project since then. He is very active in the
> > jython community,
> > a lot more than me.
> >
> > My experience has been that jnumeric is very fast indeed and can
> > really give
> > numeric a run for its money. I'm sure that whatever you do will
> > still be very
> > fast. There is really no reason to use the C/Python version, or at least,
> > there won't once you implement what you plan to implement, except for the
> > fact that the community is larger on the C/Python side.
> >
> > As far as I'm concerned though, both are excellent solutions for
> > scientific
> > programming and I use them regularly in my work. Except for the
> > fact there
> > isn't such as huge number of scientific routines available, I think that
> > Python is far superior to other solutions such as Matlab. When
> > used right,
> > Python is a great scientific tool. You can easily embed it into any
> > application later on.
> >
> > I think that a more complete numerical package for jython is of
> > great value if
> > only because everybody needs to work with matrices one day or another.
> >
> > Anyhow, you've been added as a developer.
> >
> > > My sf.net account is 'bzimmer'.  I'll add the build.xml and
> >
> > start working
> >
> > > on the LinearAlgebra package.  I noticed that some of the code is
> > > autogenerated.  What did you use to do that?
> > >
> > > thanks,
> > >
> > > brian
> > >
> > > > -----Original Message-----
> > > > From: Daniel Lemire [mailto:le...@on...]
> > > > Sent: Tuesday, December 31, 2002 4:02 PM
> > > > To: brian zimmer
> > > > Subject: Re: JNumeric LinearAlgebra
> > > >
> > > >
> > > > Sure thing. Give me your sourceforge user name and I'll add you as a
> > > > developer.
> > > >
> > > > Cheers!
> > > >
> > > > --
> > > > Daniel Lemire, Ph.D.
> > > > Research Officer, National Research Council of Canada
> > > > Adjunct Professor, University of New Brunswick
> > > > http://www.ondelette.com/acadia/
> > > >
> > > > > Hello,
> > > > >
> > > > > I recently needed the solve_linear_equations feature of
> > > >
> > > > LinearAlgebra (for
> > > >
> > > > > use with ScientificPython in Jython) and coded up a simple
> > > >
> > > > wrapper to the
> > > >
> > > > > JAMA[1] functionality.  It works well enough and is reasonably
> > > >
> > > > fast for my
> > > >
> > > > > needs.  I saw you're interested in having it fully functional
> > > >
> > > > and that you
> > > >
> > > > > have some ideas on it's implementation.  I am no mathemtician
> > > >
> > > > but a rather
> > > >
> > > > > decent developer (I'm a developer on the Jython project having
> > > >
> > > > written all
> > > >
> > > > > of the database code, zxJDBC) and I'm interested in helping out and
> > > > > brushing up on my linear algebra skills from college.  I'd be
> > > >
> > > > interested in
> > > >
> > > > > doing the work if you're willing to answer any questions I
> > > >
> > > > might have and
> > > >
> > > > > double check it for me.
> > > > >
> > > > > thanks,
> > > > >
> > > > > brian
> > > > >
> > > > > btw: You can see my current LinearAlgebra module at
> > > > > http://tinyurl.com/3t8o.
> > > > >
> > > > > btw II: I have an ant script for building JNumeric if you're
> > > > > interested.
> > > > >
> > > > > [1] http://math.nist.gov/javanumerics/jama/
> >
> > --
> > Daniel Lemire, Ph.D.
> > Research Officer, National Research Council of Canada
> > Adjunct Professor, University of New Brunswick
> > http://www.ondelette.com/acadia/

-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
Adjunct Professor, University of New Brunswick
http://www.ondelette.com/acadia/

[Jnumerical-general] Re: JNumeric LinearAlgebra

[Daniel L. <le...@on...>]

I'm not sure what you mean by "autogenerated". The stuff I have here is hand 
coded... there is no perl engine if that's what you are referring to... not 
that I know. But other people have been active on the CVS tree... Oh... maybe 
you mean the regression testing? Anyhow, there is nothing I worked on that 
was "autogenerated", we'll have to track who submitted this... 

The best thing to do at this point is join the mailing list if you haven't 
done so. There is a jnumeric mailing list on sourceforge. It has not been 
very active, but it is alive. I stronly suggest that you forward (as c.c.) 
any question or answer to that list so that future generations can benefit 
from our collective wisdom.

I was very active last year, then Frank Gibbons took over and he has been 
hacking at the project since then. He is very active in the jython community, 
a lot more than me.

My experience has been that jnumeric is very fast indeed and can really give 
numeric a run for its money. I'm sure that whatever you do will still be very 
fast. There is really no reason to use the C/Python version, or at least, 
there won't once you implement what you plan to implement, except for the 
fact that the community is larger on the C/Python side.

As far as I'm concerned though, both are excellent solutions for scientific 
programming and I use them regularly in my work. Except for the fact there 
isn't such as huge number of scientific routines available, I think that 
Python is far superior to other solutions such as Matlab. When used right, 
Python is a great scientific tool. You can easily embed it into any 
application later on.

I think that a more complete numerical package for jython is of great value if 
only because everybody needs to work with matrices one day or another.

Anyhow, you've been added as a developer. 


> My sf.net account is 'bzimmer'.  I'll add the build.xml and start working
> on the LinearAlgebra package.  I noticed that some of the code is
> autogenerated.  What did you use to do that?
>
> thanks,
>
> brian
>
> > -----Original Message-----
> > From: Daniel Lemire [mailto:le...@on...]
> > Sent: Tuesday, December 31, 2002 4:02 PM
> > To: brian zimmer
> > Subject: Re: JNumeric LinearAlgebra
> >
> >
> > Sure thing. Give me your sourceforge user name and I'll add you as a
> > developer.
> >
> > Cheers!
> >
> > --
> > Daniel Lemire, Ph.D.
> > Research Officer, National Research Council of Canada
> > Adjunct Professor, University of New Brunswick
> > http://www.ondelette.com/acadia/
> >
> > > Hello,
> > >
> > > I recently needed the solve_linear_equations feature of
> >
> > LinearAlgebra (for
> >
> > > use with ScientificPython in Jython) and coded up a simple
> >
> > wrapper to the
> >
> > > JAMA[1] functionality.  It works well enough and is reasonably
> >
> > fast for my
> >
> > > needs.  I saw you're interested in having it fully functional
> >
> > and that you
> >
> > > have some ideas on it's implementation.  I am no mathemtician
> >
> > but a rather
> >
> > > decent developer (I'm a developer on the Jython project having
> >
> > written all
> >
> > > of the database code, zxJDBC) and I'm interested in helping out and
> > > brushing up on my linear algebra skills from college.  I'd be
> >
> > interested in
> >
> > > doing the work if you're willing to answer any questions I
> >
> > might have and
> >
> > > double check it for me.
> > >
> > > thanks,
> > >
> > > brian
> > >
> > > btw: You can see my current LinearAlgebra module at
> > > http://tinyurl.com/3t8o.
> > >
> > > btw II: I have an ant script for building JNumeric if you're
> > > interested.
> > >
> > > [1] http://math.nist.gov/javanumerics/jama/

-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
Adjunct Professor, University of New Brunswick
http://www.ondelette.com/acadia/

[Jnumerical-general] More on statistical analysis and help needed for LinearAlgebra

[Daniel L. <le...@on...>]

Good day Frank,

>>I believe MLab.py is an attempt to mimic the features of MATLAB in Python. 
It is a distinct package that sits on top of Numeric.

This project intends only to make Numeric available in Jython. Maybe we 
should think about starting a new one to make MLab availabe too?<<


Hmmm... If really it "sits" on top of Numeric, why couldn't it sit on top of 
JNumeric? That's my thinking anyhow and that's why I asked him to put in a 
feature request.


However, I thought at first that we only needed ranlib... Well, ranlib is 
about 2000 lines of code in C (including all comments). It is not difficult 
code I think. I could probably do it in a week-end (I would volunteer to do 
it).

What is more of a problem is LinearAlgebra!

That's the sad truth: we need the LinearAlgebra package to support MLab...

How's that going... anyone working or thinking of working on the LinearAlgebra 
package?

Well, there are several options but we need to build LinearAlgebra on top of 
an existing package. There are several of them out there. My favorite is JSci 
because I am on the list of contributors to JSci... but I'm not sure it is 
the best choice (JSci is not performance bound even though it is quite fast).

*But* I just don't have the time to hack a LinearAlgebra package into it... at 
least not all of it myself (I could help if someone were to lead). I really 
wish we could find a volunteer... 

Anyone? ;-)

-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
http://www.ondelette.com/acadia/

[Jnumerical-general] MLab

[Daniel L. <le...@on...>]

MLab is not implemented in JNumeric.

We could easily integrate the modules MLab and RandomArray from Numeric (it is 
really a matter of "stealing" their .py files), however, RandomArray relies 
on a library called ranlib which we would need to implement in Java. It is 
not very big and could probably be hacked quickly, but it would need to be 
done by someone who knows Java and is willing to learn a bit about the 
JNumeric architecture.

Would you be so kind as to submit your feature request on JNumeric web page 
with some explanation so that we can track it?

-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
http://www.ondelette.com/acadia/

[Jnumerical-general] installer; why no statistical functions in JNumeric?

[Frank G. <fgi...@hm...>]

Daniel,

Thanks a lot for the installer. It will be useful to have something to work 
from when I need to put one together for my project.

Hassan,

JNumeric doesn't provide statistical functions. Its main emphasis has been 
on array arithmetic. However, with sum(), cumsum(), and cumproduct(), it 
should not be hard to implement the functions you need in a really compact 
and efficient way (perhaps that's why they were left out of JNumeric in the 
first place).

I believe MLab.py is an attempt to mimic the features of MATLAB in Python. 
It is a distinct package that sits on top of Numeric.

This project intends only to make Numeric available in Jython. Maybe we 
should think about starting a new one to make MLab availabe too?

-Frank

At 03:19 PM 8/27/2002, you wrote:


>I would like to perform some simple statistical operations in JNumeric (max,
>mean, median),
>and while these are fairly straightforward to write by oneself, I didn't 
>want to
>re-invent the wheel, and tried to stick with what I had been provided. I 
>saw in
>Chapter 11 of the Numerical Python tutorial that, at least under CPython, 
>there
>are some routines to allow one to do that. From the looks of things, I need to
>access scripts such as Numeric.py, MLab.py, etc. Is this the case for 
>JNumeric?
>Or are there equivalents already provided? (Simply typing commands such as
>median( a ) failed).
>
>Many thanks for your help -Hassan.

PhD, Computational Biologist,
Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115, USA.
Tel: 617-432-3555       Fax: 
617-432-3557       http://llama.med.harvard.edu/~fgibbons

[Jnumerical-general] Installer

[Daniel L. <le...@on...>]

Good day Frank,

Here's the files I promised. Nothing exciting, but it worked for my purposes. 
You can easily build something similar to download libraries as needed from a 
network. Of course, you can do *much* better than I did. What I did was a 
quick and dirty solution, but my client at the time didn't need anything 
better.

Mind you, I tried many high level tricks before going down the low and dirty 
path. They just weren't reliable enough. Mind you, I don't know about very 
recent JVMs, they may have improved (even though I doubt it).

Basically, my understanding is that if you want to avoid troubles, install 
your libraries in /jre/lib/ext. This way, you application will run as a 
(signed) applet, as a webstart application and certainly, it will work well 
with jython.

There is no reliable way (that I know of) to do this automatically. When Sun 
automatically downloads "extensions", it will put them anywhere but in 
/jre/lib/ext. That brings a lot of trouble. Plus, what happens when the 
(wrong) library is in /jre/lib/ext? Well, it gets loaded first, so no luck 
there... you really need jre/lib/ext, no matter how you turn the problem 
around (yes, you can get around that with classloaders, but then, you've got 
security issue whenever you are in a sandbox, plus, custom classloaders are 
dirty, dirty).

Mind you, even by doing it yourself, you've still got a problem: jre/ext might 
not be user writeable!  That's right! It could be that your user can't 
install extensions properly! Clearly, the classpath/jar/classloader paradigm 
is showing its age. No matter how much hacking you do, something *will* 
eventually go wrong.

In theory, for example, you could have more than one jre/lib/ext directory 
(one for each user), but Sun never provided a way to add jre/lib/ext 
directories, so you are out of luck.

Of course, you've got also the problem that extensions have no reliable 
versioning. That means that you may or may not write over an extension needed 
by another application (that happens often with XML parser), you may 
incorrectly assume that the proper extension is in place whereas you've 
really got an older or newer version in place... and so on. So, you would 
need to check this, check that... ah...

Basically, it is very, very dirty. All these things will never get fixed I'm 
afraid. Java is a very nice language... but its hasty beginnings are a major 
drag on it. And any time someone suggest to break away from the old tricks 
and evolve into something better, they typically get hammered so hard that 
nobody dares to speak out loud about Java shortcomings anymore... nobody from 
the Java community, that is.



-- 
Daniel Lemire, Ph.D.
Research Officer, National Research Council of Canada
http://www.ondelette.com/acadia/