OCaml App (NML) Announce

OCaml App (NML) Announce

[David McClain]

Dear OCaml Enthusiasts,

It has been stewing for more than a year now, a continuing work in progress,
but it is high time that I release a matured copy of the code and sources to
the world. NML (Not ML, Numeric Modeling Language, Numeric ML, Nearly ML,
...) is an interactive, dynamically typed, tail pure, compiled (to native
code closures) functional language, whose syntax closely follows that of
OCaml, but where all math operations are overloaded and vectorized on real
and complex data in the form of lists, vectors, multidimensional arrays,
tuples, etc.

It has proven itself in the field for the past 9 months. Numerous samples
are included with the sources, including a translation of Norvig's Prolog
interpreter (just a toy... but it shows the power of NML for non-numeric as
well as numeric problems). NML is very fast!!! on large array-based
problems, and is reasonably fast on non-numeric problems (probably not as
efficient as OCaml) but certainly a lot easier to code interactively at the
command line (no type inferencing and no type checking... hence inherently
unsafe).

The application and its sources presently runs on Win/NT 4.0 and Linux. But
the Linux port has been ignored for the past 5 months. It produces very nice
looking graphics, 2-D data plots, pseudo-color image displays, and shaded
surface plots. It is shareware in the sense of the OCaml license, and a
request that acknowledgement be given to the original authors. Source
consists of about 28K lines of OCaml, and 10K lines of supporting C/C++
code.

You can find more about it at
http://www.azstarnet.com/~dmcclain/nmlpromo.html
and the zipped sources and NML.exe at
http://www.azstarnet.com/~dmcclain/nml.zip  (1100 KB).

Many thanks to Xavier and the others at INRIA for their wonderful language
system!!

- D. McClain, Sr. Scientist
Raytheon Systems Co.
Tucson, AZ

scientific computing with ocaml, gsl api

[Julian Assange]

"David McClain" <dmcclain@azstarnet.com> writes:

> Dear OCaml Enthusiasts,
> 
> It has been stewing for more than a year now, a continuing work in progress,
> but it is high time that I release a matured copy of the code and sources to
> the world. NML (Not ML, Numeric Modeling Language, Numeric ML, Nearly ML,
> ...) is an interactive, dynamically typed, tail pure, compiled (to native
> code closures) functional language, whose syntax closely follows that of
> OCaml, but where all math operations are overloaded and vectorized on real
> and complex data in the form of lists, vectors, multidimensional arrays,
> tuples, etc.

This looks very nice david! Is it possible to use the vectorised, array support
within ocaml? i.e I'm a little leary of using NML for mid-large applications due
to the lack of type checking, but it does seem to be an excellent language for
scientific interrogation.

Have you looked at the GNU scientific library?

        http://sourceware.cygnus.com/gsl

This is a wonderfully eclectic scientific library in C, with strong
control over float properties. An ocaml or MNL binding would be a
killer app.

 > Are there any plans to support euclidian vector algebra in n
 > dimensions? Preferably with user-defined physical field properties?
 > 
 > Specifically I want to be able to do things like define two vectors,
 > v_1, and v_2, have v_1 radiate a force decreasing at 1/distance^2, and
 > calculate the the force vector across all of v_2. This is more complex
 > than simple point sources, but there doesn't even seem to be support
 > for those. It could be argued that a two body case is so trivial it
 > doesn't need supporting, which is probably true, but n body cases and
 > non point sources are hard work and useful in many (even non-physics)
 > applications. i.e the v_1, v2 example I mentioned above forms part of
 > an optimisation solution I have for laying out 2d chemical labels
 > (part-of-molecule number, atomic weight, charge, etc) over a 3d
 > polynucleartide in such a way as to avoid the labels writing accross
 > each other. 
 >
 > Cheers,
 > Julian

Re: scientific computing with ocaml, gsl api

[David McClain]

Actually, there are OCaml bindings available in NML for just this purpose.
My first major application of NML was a hybrid OCaml/NML program for solving
a highly nonlinear problem that computes the phase variations in an optical
system that give rise to measured image blur functions. If you are
interested in a copy of the source for this app (about 700 lines of OCaml/50
lines of NML) just drop me a note. The OCaml is terrific for overall program
organization, while the NML is quite expressive for the array computations
and Fourier analysis...

- DM

----- Original Message -----
From: Julian Assange <proff@iq.org>
To: David McClain <dmcclain@azstarnet.com>
Cc: <caml-list@inria.fr>; <proff@iq.org>
Sent: Tuesday, March 28, 2000 11:00 PM
Subject: scientific computing with ocaml, gsl api


> "David McClain" <dmcclain@azstarnet.com> writes:
>
> > Dear OCaml Enthusiasts,
> >
> > It has been stewing for more than a year now, a continuing work in
progress,
> > but it is high time that I release a matured copy of the code and
sources to
> > the world. NML (Not ML, Numeric Modeling Language, Numeric ML, Nearly
ML,
> > ...) is an interactive, dynamically typed, tail pure, compiled (to
native
> > code closures) functional language, whose syntax closely follows that of
> > OCaml, but where all math operations are overloaded and vectorized on
real
> > and complex data in the form of lists, vectors, multidimensional arrays,
> > tuples, etc.
>
> This looks very nice david! Is it possible to use the vectorised, array
support
> within ocaml? i.e I'm a little leary of using NML for mid-large
applications due
> to the lack of type checking, but it does seem to be an excellent language
for
> scientific interrogation.
>
> Have you looked at the GNU scientific library?
>
>         http://sourceware.cygnus.com/gsl
>
> This is a wonderfully eclectic scientific library in C, with strong
> control over float properties. An ocaml or MNL binding would be a
> killer app.
>
>  > Are there any plans to support euclidian vector algebra in n
>  > dimensions? Preferably with user-defined physical field properties?
>  >
>  > Specifically I want to be able to do things like define two vectors,
>  > v_1, and v_2, have v_1 radiate a force decreasing at 1/distance^2, and
>  > calculate the the force vector across all of v_2. This is more complex
>  > than simple point sources, but there doesn't even seem to be support
>  > for those. It could be argued that a two body case is so trivial it
>  > doesn't need supporting, which is probably true, but n body cases and
>  > non point sources are hard work and useful in many (even non-physics)
>  > applications. i.e the v_1, v2 example I mentioned above forms part of
>  > an optimisation solution I have for laying out 2d chemical labels
>  > (part-of-molecule number, atomic weight, charge, etc) over a 3d
>  > polynucleartide in such a way as to avoid the labels writing accross
>  > each other.
>  >
>  > Cheers,
>  > Julian
>

Re: scientific computing with ocaml, gsl api

[David McClain]

Julian,

I'm not quite sure what you mean by Euclidean N-dimensional algebra, but NML
does already support inner and outer products of N-dimensional objects.
There are also Matrix routines for SVD, Cholesky decomposition, and LU
decomposition, inversion, and determinants, etc...

In fact one of the things I do quite frequently is a DFT demodulation of
image stacks obtained with a chopped sensor. Hence to multiply and sum the
image planes with a twiddle vector I simply do:

    images <*> twiddles

and voila, the result is a single image that represents the sum of
individual weighted images. The "images" vector above is a 3-dimensional
stack of 2-D image planes, while the "twiddles" is a simple vector of scalar
(complex) weights. The "<*>" is shorthand for my inner-product operator.

I am sure that you will easily represent your force problem in the
vectorized domain of NML!

- DM

----- Original Message -----
From: Julian Assange <proff@iq.org>
To: David McClain <dmcclain@azstarnet.com>
Cc: <caml-list@inria.fr>; <proff@iq.org>
Sent: Tuesday, March 28, 2000 11:00 PM
Subject: scientific computing with ocaml, gsl api


> "David McClain" <dmcclain@azstarnet.com> writes:
>
> > Dear OCaml Enthusiasts,
> >
> > It has been stewing for more than a year now, a continuing work in
progress,
> > but it is high time that I release a matured copy of the code and
sources to
> > the world. NML (Not ML, Numeric Modeling Language, Numeric ML, Nearly
ML,
> > ...) is an interactive, dynamically typed, tail pure, compiled (to
native
> > code closures) functional language, whose syntax closely follows that of
> > OCaml, but where all math operations are overloaded and vectorized on
real
> > and complex data in the form of lists, vectors, multidimensional arrays,
> > tuples, etc.
>
> This looks very nice david! Is it possible to use the vectorised, array
support
> within ocaml? i.e I'm a little leary of using NML for mid-large
applications due
> to the lack of type checking, but it does seem to be an excellent language
for
> scientific interrogation.
>
> Have you looked at the GNU scientific library?
>
>         http://sourceware.cygnus.com/gsl
>
> This is a wonderfully eclectic scientific library in C, with strong
> control over float properties. An ocaml or MNL binding would be a
> killer app.
>
>  > Are there any plans to support euclidian vector algebra in n
>  > dimensions? Preferably with user-defined physical field properties?
>  >
>  > Specifically I want to be able to do things like define two vectors,
>  > v_1, and v_2, have v_1 radiate a force decreasing at 1/distance^2, and
>  > calculate the the force vector across all of v_2. This is more complex
>  > than simple point sources, but there doesn't even seem to be support
>  > for those. It could be argued that a two body case is so trivial it
>  > doesn't need supporting, which is probably true, but n body cases and
>  > non point sources are hard work and useful in many (even non-physics)
>  > applications. i.e the v_1, v2 example I mentioned above forms part of
>  > an optimisation solution I have for laying out 2d chemical labels
>  > (part-of-molecule number, atomic weight, charge, etc) over a 3d
>  > polynucleartide in such a way as to avoid the labels writing accross
>  > each other.
>  >
>  > Cheers,
>  > Julian
>

Re: scientific computing with ocaml, gsl api

[David McClain]

It would appear that any external library developed in C would be a huge
risk to code safety. NML was coded to the greatest possible degree in OCaml
for exactly this reason. It (OCaml, and hence NML) has a superb and robust
GC, and runtimes errors cannot crash the apps. Instead, if an NML routine is
misused, it simply gives an error message at runtime, along with a traceback
of most recent call paths, identify source file and line from which the
error propagated.

I am personally leary of C code, even though I have more than 20 years
experience with it -- perhaps because I have so much experience with it! In
every recent case where a program exhibited anomalous behavior and needed
debugging I could trace it back to the use of C/C++. These are generally
off-by-one boundary condition errors, but sometimes logic errors that have
been obfuscated by the lack of clean syntax such as offered by OCaml. I
don't generally have dangling pointer problems in my life, but many others
do. I personally prefer to have a robust and fast GC running so I don't have
to become a memory accountant.

Hence, if the "gsl" library were to be recast entirely (or almost so) in
OCaml I would be much more interested in using it. As it stands in C, I
would be quite distrustful of it.

- DM


----- Original Message -----
From: Julian Assange <proff@iq.org>
To: David McClain <dmcclain@azstarnet.com>
Cc: <caml-list@inria.fr>; <proff@iq.org>
Sent: Tuesday, March 28, 2000 11:00 PM
Subject: scientific computing with ocaml, gsl api


> "David McClain" <dmcclain@azstarnet.com> writes:
>
> > Dear OCaml Enthusiasts,
> >
> > It has been stewing for more than a year now, a continuing work in
progress,
> > but it is high time that I release a matured copy of the code and
sources to
> > the world. NML (Not ML, Numeric Modeling Language, Numeric ML, Nearly
ML,
> > ...) is an interactive, dynamically typed, tail pure, compiled (to
native
> > code closures) functional language, whose syntax closely follows that of
> > OCaml, but where all math operations are overloaded and vectorized on
real
> > and complex data in the form of lists, vectors, multidimensional arrays,
> > tuples, etc.
>
> This looks very nice david! Is it possible to use the vectorised, array
support
> within ocaml? i.e I'm a little leary of using NML for mid-large
applications due
> to the lack of type checking, but it does seem to be an excellent language
for
> scientific interrogation.
>
> Have you looked at the GNU scientific library?
>
>         http://sourceware.cygnus.com/gsl
>
> This is a wonderfully eclectic scientific library in C, with strong
> control over float properties. An ocaml or MNL binding would be a
> killer app.
>
>  > Are there any plans to support euclidian vector algebra in n
>  > dimensions? Preferably with user-defined physical field properties?
>  >
>  > Specifically I want to be able to do things like define two vectors,
>  > v_1, and v_2, have v_1 radiate a force decreasing at 1/distance^2, and
>  > calculate the the force vector across all of v_2. This is more complex
>  > than simple point sources, but there doesn't even seem to be support
>  > for those. It could be argued that a two body case is so trivial it
>  > doesn't need supporting, which is probably true, but n body cases and
>  > non point sources are hard work and useful in many (even non-physics)
>  > applications. i.e the v_1, v2 example I mentioned above forms part of
>  > an optimisation solution I have for laying out 2d chemical labels
>  > (part-of-molecule number, atomic weight, charge, etc) over a 3d
>  > polynucleartide in such a way as to avoid the labels writing accross
>  > each other.
>  >
>  > Cheers,
>  > Julian
>