Neuroimage software - scientific computing and visualization

Ruby
1

Hi,

I’ve been programming with perl, a bit of python and a lot of matlab.
I’ve become very curious about ruby and considering whether or not to
take it on as my main development environment in the near future. I
could use some good advice before I dive into the deep end.

I’m looking at Ruby for scripting/programming of scientific computing
for neuroimaging. This includes time-series plots and analysis, image
volume segmentation and rendering, surface modeling, visualization and
morphing, and animating scalar and vector quanities on surfaces. Also
linear algrebra methods and spatial transformations, almost any
scientific computation you can imagine, especially large scale
statistical analysis (including permutations). Did I leave anything
out? What about threading and distributed cluster systems? I could
really use some tips on whether Ruby is a good choice for this work and
where to find projects that already fit into this picture.

So, I’m wonder if ruby has wrapped things like: VTK, ITK, The R-project
for stats, OpenGL, wxWindows, scientific libraries (eg, GNU GSL,
Atlas/Lapack), graphics libraries, etc. Any good advice on getting
started with these tools would be great.

There is some momentum in this area using python already. Why should I
use Ruby instead of Python? One thing that I’ve noticed is that python
distribution and installation methods are clumsy. If I want some
python package for neuroimaging, I have to find and install many
additional dependencies before I even get started. As a Debian user,
the apt system helps, but it’s still not quite what I want. Ruby has
the gem system that appears to do the job right.

Thanks, Darren

2

I think you should stick with Matlab, they have developped programs for
years for this kind of work. The programs I have seen for MRI imaging
are in matlab.

In ruby, there are things missing: ex when you do a sort, you don’t get
the index of a sort as in matlab, only the sorted array. So you would
have to write you own - Matlab like - methods.

The other thing I miss in ruby, the ability to do matrix operations. If
you do the mean, it is the mean of a vector, while in matlab, you can
do a mean on any dimension array, indicating which dimension

I don’t know how much stats ruby can do, certainly the matlab stat
package is not complete, but I can’t imagine that a programming
language would be that developed in this area.

Another reason to stay with matlab is readability. Matlab is a higher
level language. It sounds like you are a programmer selling a product,
so it might be different, but I work at a university, and students
inherit the matlab programs from one another, and then typically they
have to figure out what the program does to get the data in correctly
for their data, get the figures they want for their conditions… the
fact that it is written in a higher level language seems to help.

Ruby does have some matrix computation ability, so you should look at
that, and see if that suits your needs.

3

Hi Darren

I’m looking at Ruby for scripting/programming of scientific computing
for neuroimaging.

I used Matlab for my PhD research (medical image modelling and
analysis), and I think it was excellent for my purposes. Below is a
list of why I liked using it, which I’ll compare to my (limited)
experience with Ruby afterwards:

  1. Matlab language syntax maps closely to the underlying concepts (i.e.
    vector and matrices).
  2. Very simple, but powerful, plotting/visualisation abilities.
  3. (Generally) efficient implementations for common matrix operations
    (multiplication, decompositions etc.). These are very well-tested.
  4. A REPL (i.e. you can experiment on a ‘command line’).
  5. Cross-platform (Linux, OS X, Solaris, Windows).
  6. No compiling (recent Matlabs do just-in-time compilation
    automatically).
  7. Excellent support for profiling (easily find and quantify those
    computational hot-spots).
  8. It’s easy to write C (or FORTRAN) extensions when you really need
    every last ounce of speed or control over memory.
  9. Lots of other people use it, so there’s an ‘ecosystem’ (the users
    are also working on your problem or closely-related ones).

Comparing this to Ruby:

  1. Ruby is high-level, but it’s a general purpose language; it’s not
    specialised for vectors and matrices. You’d have to get a library or
    write this yourself. The Ruby syntax is ‘sort-of’ similar to the Matlab
    syntax, though. Well, it feels Matlab-y to me!
  2. You’d need to write or find plotting/visualisation library for Ruby.
    I’ve not looked, but I’d hope this is reasonably trivial.
  3. You’d need to find or write a library to let you do efficient matrix
    computations. I know there is stuff for Ruby out there, but it won’t be
    as well-tested as Matlab’s and it’s not a first class citizen of the
    environment. Writing your own is much harder than it seems. You really
    need a library that allows you to call LAPACK routines from Ruby.
  4. Ruby has the irb REPL. This is fine.
  5. Ruby seems to be pretty cross-platform, but you’d need to check the
    same is true for the additional libraries you’d need. You might not
    care about WIndows, but what about your colleagues? You often need to
    share code with them.
  6. No compiling, but also no JIT as far as I’m aware. Not too big an
    issue if you’ve got a good matrix library.
  7. I’ve not tried profiling in Ruby, but Matlab’s profiler is very
    good. If you can’t profile your code and you’re doing numerics, you’re
    in trouble.
  8. I understand you can write Ruby extensions in C but I’ve not done
    it.
  9. The ecosystem for your problem domain is going to be much smaller in
    Ruby than in Matlab. Ruby’s best ecosystem domain seems to be webapps
    (i.e. Rails).

Now, the thing I haven’t mentioned is price. Matlab is expensive
(thousands of dollars for all the stuff you need). They do student
discounts, but last time I looked you’d need to pay hundreds of dollars
rather than thousands, so it’s not a great discount. (Mathematica’s
student terms were much better, last time I looked.)

Ruby is free. In my opinion, if you can afford Matlab (or someone else
is affording it for you), you should have a very good reason before you
ditch Matlab. However, if you can’t afford Matlab, or have ideological
reasons why you don’t want to use it (it’s not open source, for
example), you should probably use and improve Octave, the open source
equivalent to Matlab. However, if you need to roll your own soluation,
I think Ruby would make a very good foundation for putting together
your own environment for numerical computing.

I think the real question is whether you want to do the ‘fun’ thing and
start (almost) from scratch (i.e. base your environment on Ruby) or
stand on the shoulders of giants (i.e. use Matlab or Octave). If you’ve
got real problems to solve now, you should probably resist the
temptation to do the fun thing.

Chris

4

In article [email protected],
Darren L. Weber [email protected] wrote:

volume segmentation and rendering, surface modeling, visualization and
Atlas/Lapack), graphics libraries, etc. Any good advice on getting
started with these tools would be great.

wxWindows, GSL, OpenGL are there, I believe. Not sure about the rest.

It would be nice to get a nice Ruby->R bridge and it could probably be
very Rubyish (maybe a DSL built on Ruby).

There is some momentum in this area using python already. Why should I
use Ruby instead of Python? One thing that I’ve noticed is that python
distribution and installation methods are clumsy.

Well, that could be your reason. The other thing to consider is which
language
you prefer. While many people will claim they are equivilent, there are
differences both feature-wise and less tangible differences. Check out
Bruce
Eckel’s recent post at Artima, for example.

If I want some
python package for neuroimaging, I have to find and install many
additional dependencies before I even get started. As a Debian user,
the apt system helps, but it’s still not quite what I want. Ruby has
the gem system that appears to do the job right.

You should definitely check ou the SciRuby site:
http://sciruby.codeforpeople.com/

You should also check out narray for high speed vector math in Ruby.

Phil

5

In article [email protected],
anne001 [email protected] wrote:

I think you should stick with Matlab, they have developped programs for
years for this kind of work. The programs I have seen for MRI imaging
are in matlab.

In ruby, there are things missing: ex when you do a sort, you don’t get
the index of a sort as in matlab, only the sorted array. So you would
have to write you own - Matlab like - methods.

I’m not sure I understand what you’re saying here: are you saying you
want the
indices of the array sorted, but not the array itself?

So are you saying that if you have an array like:
a = [‘d’, ‘b’, ‘a’, ‘c’, ‘e’]
You would like to get something like:
a.sort_index #=> [2, 1, 3, 0, 4]

If that’s the case, then I’m sure it’s doable. No built-in method that
I’m
aware of, but you could easily create one.

The other thing I miss in ruby, the ability to do matrix operations. If
you do the mean, it is the mean of a vector, while in matlab, you can
do a mean on any dimension array, indicating which dimension

check out narray.

for their data, get the figures they want for their conditions… the
fact that it is written in a higher level language seems to help.

YMMV. I don’t find Matlab very readable. I much prefer Ruby’s syntax.

It’s been mentioned before, but it would be cool to create a kind of
MatLab
‘replacement’ using Ruby. One could imagine that you could take
advantage of
Ruby’s metaprogramming and DSL abilities to create something that would
be
quite nice to use. (realistically, we couldn’t completely replicate all
matlab
functionality, but perhaps we could focus on certain areas).

MatLab may be ‘higher level’ but that’s only in the sense of
representing
mathematical operations. In another sense Ruby is higher-level because
it
would allow you to create a DSL for representing mathematical operations
(similar to MatLab).

Phil

6

Thanks very much for these thoughtful comments.

Matlab is very nice and I enjoy working with it. The help system is
very informative and all aspects of scientific computing work well. No
wonder so many university projects work with matlab. But, I do not
enjoy the inconvenience of running a licence manager and having my work
come to a halt when my license expires every year. (Yes, some IT
departments do not automatically update the licenses on every system,
including my laptop). I also fear that sometime I may not have a
matlab license and my work will be locked, despite my best attempts to
create open-source tools. Most annoying is that I do not know how to
or cannot create multi-threaded functions and I can’t call matlab
functions in shell scripts (matlab doesn’t have a shebang for
scripting).

I can see some momentum in python for scientific computing and I have
begun to play with python, but I find the installation of modules can
be either easy or painful (especially dependencies). My intuitions are
telling me that ruby could be a better option than python in the long
term, but I want to clarify my understanding before I get carried away
with enthusiasm.

For a start, I’ve noticed these useful sites:

http://narray.rubyforge.org/index.html.en

http://rb-gsl.rubyforge.org/

http://sciruby.codeforpeople.com/sr.cgi/FrontPage

I don’t see anyone using these for neuroimaging, in particular.

Best, Darren

7

In article [email protected],
[email protected] wrote:

  1. Matlab language syntax maps closely to the underlying concepts (i.e.
  2. It’s easy to write C (or FORTRAN) extensions when you really need
  3. You’d need to write or find plotting/visualisation library for Ruby.
    I’ve not looked, but I’d hope this is reasonably trivial.

There are a few things out there. There’s a gnuplot interface, for
example.

  1. You’d need to find or write a library to let you do efficient matrix
    computations. I know there is stuff for Ruby out there, but it won’t be
    as well-tested as Matlab’s and it’s not a first class citizen of the
    environment. Writing your own is much harder than it seems. You really
    need a library that allows you to call LAPACK routines from Ruby.

Seems to me that narray has most of what you need here.

  1. Ruby has the irb REPL. This is fine.
  2. Ruby seems to be pretty cross-platform, but you’d need to check the
    same is true for the additional libraries you’d need. You might not
    care about WIndows, but what about your colleagues? You often need to
    share code with them.

No, they’ve all bought iBooks or PowerBooks by now :wink:

  1. No compiling, but also no JIT as far as I’m aware. Not too big an
    issue if you’ve got a good matrix library.

narray is implemented in C.

  1. I’ve not tried profiling in Ruby, but Matlab’s profiler is very
    good. If you can’t profile your code and you’re doing numerics, you’re
    in trouble.

Ruby has reasonably good profiling ability.

  1. I understand you can write Ruby extensions in C but I’ve not done
    it.

It’s quite easy. It’s even easier if you use Ruby::Inline.

Here’s an tantalizing idea that I’ll throw out: C is one thing and,
yes, it’s
fast… but what if you could actually map some algorithm or math
operation to
hardware and thus get a large speed advantage even over C? I’m
working on
something like this… details forthcoming. The idea is very similar to
Ruby::Inline. It will allow seamless communication between your Ruby
program
and an FPGA board acting as hardware accelerator (these boards are
available
for less than $200 now).

  1. The ecosystem for your problem domain is going to be much smaller in
    Ruby than in Matlab. Ruby’s best ecosystem domain seems to be webapps
    (i.e. Rails).

Maybe that’s the most successful domain related to Ruby up to this
point,
however there are people using Ruby for scientific apps - see the
SciRuby site:
http://sciruby.codeforpeople.com/

Ruby isn’t just for webapps. In fact, until the last 18 months or so,
people
were wondering if Ruby would be a player in the webapp space - how
quickly
things change, eh?

reasons why you don’t want to use it (it’s not open source, for
example), you should probably use and improve Octave, the open source
equivalent to Matlab. However, if you need to roll your own soluation,
I think Ruby would make a very good foundation for putting together
your own environment for numerical computing.

Yes, and part of the philosophical argument as I see it is that tools
like
MatLab are becoming ‘required’ for doing research (many people think so
anyway), kind of like using LaTeX for writing papers. However, LaTeX is
free.
It would be great to have a completely free tool for research since
computational methods are becoming central to so much research.

I think the real question is whether you want to do the ‘fun’ thing and
start (almost) from scratch (i.e. base your environment on Ruby) or
stand on the shoulders of giants (i.e. use Matlab or Octave). If you’ve
got real problems to solve now, you should probably resist the
temptation to do the fun thing.

However, I think you’d be surprised at how much is already available in
Ruby.
Check it out and see if what you need might already be there (or be
almost
there). Also, get on SciRuby and post requests.

Phil

8

On Fri, 3 Feb 2006 [email protected] wrote:

functions in shell scripts (matlab doesn’t have a shebang for

Thanks for the pointers.

Chris

i do a lot with narray, mmap, and gsl. let me know if you have
questions.

what are your current projects? if you don’t mind me asking…

cheers.

-a

9

Hi Darren

But, I do not
enjoy the inconvenience of running a licence manager and having my work
come to a halt when my license expires every year. (Yes, some IT
departments do not automatically update the licenses on every system,
including my laptop). I also fear that sometime I may not have a
matlab license and my work will be locked, despite my best attempts to
create open-source tools. Most annoying is that I do not know how to
or cannot create multi-threaded functions and I can’t call matlab
functions in shell scripts (matlab doesn’t have a shebang for
scripting).

All valid points, except that you can call matlab functions on the UNIX
(and Windows?) command line (though probably not with a shebang, but I
don’t see that as essential).

For a start, I’ve noticed these useful sites:
http://narray.rubyforge.org/index.html.en
http://rb-gsl.rubyforge.org/
http://sciruby.codeforpeople.com/sr.cgi/FrontPage

Thanks for the pointers.

Chris

10

On Fri, 3 Feb 2006 [email protected] wrote:

Agreed.
but you can use many gsl methods with narray - check out the docs.

narrays are also stored in memory just like c arrays - so you may call a
whole
world of methods on the data via third party libs or even dl calls.

cheers.

-a

11

Phil

narray is implemented in C.

I took a quick look at its API and it only seemed to offer the LU
matrix decomposition; which seems a bit minimal.

It would be great to have a completely free tool for research since
computational methods are becoming central to so much research.

Agreed.

Chris

12

Phil T. wrote:

In article [email protected],
anne001 [email protected] wrote:
…

fact that it is written in a higher level language seems to help.

YMMV. I don’t find Matlab very readable. I much prefer Ruby’s syntax.

Unless MATLAB has changed in the last 3 or 4 years, it’s a real pain to
work with objects, as opposed to numbers. The problem I could not
overcome was that all assignments have copy semantics rather than
reference semantics. There was no way to take a reference to an object
and assign that to a variable or to a field in some structure. So is was
effectively impossible to build data structures whose reference graphs
were not trees.

13

Phil T. wrote:

Here’s an tantalizing idea that I’ll throw out: C is one thing and, yes, it’s
fast… but what if you could actually map some algorithm or math operation to
hardware and thus get a large speed advantage even over C? I’m working on
something like this… details forthcoming. The idea is very similar to
Ruby::Inline. It will allow seamless communication between your Ruby program
and an FPGA board acting as hardware accelerator (these boards are available
for less than $200 now).

That would be very cool. Eventually, we’re going to be porting a
wireless network stack to FPGA, and being able to still use Ruby would
be very nice.

What kind of code goes inside your “inline” blocks?

14

Acceleration of algorithms by moving the number-crunching parts to FPGAs
has been well-utilized in the life sciences industry for a while now.
Due to non-disclosure agreements, I cannot provide the details, but my
company has been providing such solutions for the last couple of years.

Normally, once the board design is finalized, a suitable C
implementation of the algorithm is used as the basis for a Verilog
implementation to be burned into the FPGA. Typically, these FPGAs are
mounted on PCI cards or USB devices.

A driver does the interface between the application in software and the
FPGA, handling interrupt events, data handover, etc.

I would be interested in your work, unless you are building a commercial
closed-source solution.

Best regards,

JS

15

In article [email protected],
Joel VanderWerf [email protected] wrote:

wireless network stack to FPGA, and being able to still use Ruby would
be very nice.

What kind of code goes inside your “inline” blocks?

Initially it’s going to be VHDL, a hardware description language. There
are
lots of efforts afoot to use C and a hardware description language, but
I’m
just not sure yet that there’s much advantage (YET) over using an HDL
like
VHDL. The semantics of an HDL are much different than the semantics of
a
programming language like C or Ruby. HDLs have to model the parallelism
inherant in hardware.

So the idea of RubyInline::HDL is evolutionary rather than
revolutionary: use
Ruby for what it’s good for and use HDLs for what they’re good for. Of
course,
from a practical standpoint, using VHDL also makes it easy to use the
free
(though not open source, unfortuneatly) tools from Xilinx to program
their
FPGAs.

In the future, I’d like to build on RHDL (Ruby as an HDL) so that it can
generate VHDL as an output. At that point then the inlined ‘language’
would
be RHDL (really a Ruby-based DSL). Also, at some point it would be good
to
support FpgaC which is an open source C programming environment for
FPGAs.

In the short term, you’ll either need to know VHDL to use Inline::HDL,
or
perhaps we could build a repository (using gems perhaps?) of useful VHDL
modules that users could download from some website. Lots of common DSP
functions could be made available. This could make it so that you
wouldn’t
have to know much VHDL to use Inline::HDL.

So, for example, say you want to use Ruby to do some image processing.
Maybe
you want that image processing to be in real time. Maybe C isn’t even
fast
enough for whatever you’re doing. You download some image processing
packages
which are written in VHDL and you inline them using Inline::HDL. Your
system
would look something like:

camera → FPGA (running image processing algorithms)->PCI bus->Ruby
program

In this scenario the dataflow is pretty much one way, but you can
imagine other
scenarios where your Ruby program sends some data to the FPGA to be
processed
and waits for an answer back. Maybe you’re doing some sort of
optimization
that would take a long time in software, but would be much faster if it
can be
implemented in hardware. Your Ruby program would send a block of data
over the
PCI bus to the FPGA card, the FPGA card would then process that data and
send
back the processed data.

RubyInline::HDL would take care of setting up all the interfaces
(hardware and
software) between the Ruby program and the FPGA board and it would take
care
of running the Xilinx tools to create the bitstream (if one doesn’t
already
exist). That’s the theory anyway; this is all very preliminary based on
some
work I did last summer using a
VHDL simulator (ghdl) in place of an FPGA board. I need to finish that
up to
get testing going and then start working with actual hardware. Now
that my Masters degree is about done (defense is tomorrow!) I’ll have
some time
to put into this over the next couple of months.

Phil

16

On Fri, 3 Feb 2006, Joel VanderWerf wrote:

wireless network stack to FPGA, and being able to still use Ruby would
be very nice.

What kind of code goes inside your “inline” blocks?

this sounds insanely cool. can you ping me offline phil?

-a

17

In article [email protected],
Srinivas J. [email protected] wrote:

Acceleration of algorithms by moving the number-crunching parts to FPGAs
has been well-utilized in the life sciences industry for a while now.
Due to non-disclosure agreements, I cannot provide the details, but my
company has been providing such solutions for the last couple of years.

Yes, this seems to be going on, though it’s kind of been in the
background.

Normally, once the board design is finalized, a suitable C
implementation of the algorithm is used as the basis for a Verilog
implementation to be burned into the FPGA. Typically, these FPGAs are
mounted on PCI cards or USB devices.

Right, the idea here would be to either use the Raggedstone boards or
one of
the newer Xilinx boards which has USB. The architectures needs to be
modular
to support different boards, though initially I’ll concentrate on Xilinx
FPGAs.

A driver does the interface between the application in software and the
FPGA, handling interrupt events, data handover, etc.

Right, that’s what I’m thinking, though I want to make all of that
interface
stuff as transparent as possible to the user by using Ruby to generate
code,
etc.

I would be interested in your work, unless you are building a commercial
closed-source solution.

The idea is for this to be open source. But if someone wants to pay me
for
support, such as developing the VHDL code, etc. that’s cool. I’m
actually
hoping that as this sort of thing gets popular that there could be some
way for
me to derrive some income from this. I’m planning to spend the next
three
months or so on this putting in 20 to 40 hours/week, so I hope for some
kind of payoff - it’s kind of going to be my unpaid job unless I run out
of
money and have to look for more paying work :wink:

I think that FPGA acceleration is
the way we’re going to move to the next level of performance as general
purpose
CPUs are not increasing in performance as much as they used to.

Phil

18

On 2/2/06, Phil T. [email protected] wrote:

In the future, I’d like to build on RHDL (Ruby as an HDL) so that it can
generate VHDL as an output. At that point then the inlined ‘language’ would
be RHDL (really a Ruby-based DSL). Also, at some point it would be good to
support FpgaC which is an open source C programming environment for FPGAs.

I really don’t know anything about actual (as opposed to armchair
quarterback) hardware design, but I would have sworn one of the
presentations at RubyConf 2005 included a brief mention of a
domain-specific language for hardware. I remember something about
circuit board layouts/layers/etc.
I thought it was in Jim W.'s DSL talk, but I just ran through the
presentation slides, and didn’t see it. Maybe it was something he
pulled up on his laptop, outside of the presentation?

–Wilson

19

Phil T. wrote:

It would be nice to get a nice Ruby->R bridge and it could probably be
very Rubyish (maybe a DSL built on Ruby).

There is a Ruby wrapper for the core R math library (Rmath) on RAA. I
got it working a few weeks ago but had to drop the project in favor of
something I get paid to do (with R, however). I hate to show up on a
Ruby list as an R bigot, but, well, I am. :slight_smile:

Specifically for neuroimaging, one would need a Ruby wrapper for the
BLAS and LAPACK libraries and probably wfft. I don’t think either is
particularly difficult to do, given the excellent instructions on the
Ruby/C interface in the Pickaxe book. I just personally have little
motivation to do it.

By the way, if you’re going to use BLAS and LAPACK, go the extra mile
and use the Automatically Tuned Linear Algebra Subroutines (ATLAS) as
well. You’ll get near machine speed on decent-sized problems.

–
M. Edward (Ed) Borasky

20

Phil T. wrote:

Yes, and part of the philosophical argument as I see it is that tools like
MatLab are becoming ‘required’ for doing research (many people think so
anyway), kind of like using LaTeX for writing papers. However, LaTeX is free.
It would be great to have a completely free tool for research since
computational methods are becoming central to so much research.

If you want to be part of a research community, you need a good reason
not to use the language that community uses! In the finance community,
APL, Matlab, Mathematica and to some extent R are the tools of choice.
In digital signal processing, if you don’t know Matlab, you might as
well find some other line of work. In statistics, there are more
options, but R is a heavyweight, as it is in bioinformatics.

But let’s talk briefly about money. :slight_smile: One of the things that still
amazes me, even though I’ve been part of a few open source communities
for almost six years now, is the quality of free/open source software,
essentially built by people who are doing it for love and not for money.
There are four open source LISP implementations, Perl, Python, PHP,
Ruby/Rails/PostgreSQL/MySQL/SQLite, R (and all the R library packages),
Maxima, Axiom, Atlas, TeXmacs, LyX, and of course KDE, sitting on my
workstation right now. And that workstation is running the free Gentoo
Linux OS, the Linux 2.6 kernel and the entire GNU toolchain. And it all
works – most of it beta-level and sometimes even alpha-level!

I haven’t added up the lines of code – which I could do easily since
Gentoo is built from source :slight_smile: – but it must be in the millions! To
collect the functional equivalents in commercial software … well, for
openers, the entry-level price for Matlab and Mathematica alone is about
$1000 US each, IIRC. Add in SQL Server 2000, Windows XP, Visual Studio,
a decent scientific typesetting package, a desktop publishing package,
etc., and you’ll probably spend $10000 US on software to get what I have

One thing Ruby (and Perl and Python) is very good at is providing the
“glue logic” to build upon existing software, whether free/open source
or commercial. The PragProg folks even have a book on the subject, IIRC
called “Enterprise Integration with Ruby”. So … there is a free tool
for just about every research need, and there’s no reason in the world a
Ruby programmer can’t have the best of all worlds.

</Saturday morning rant>

–
M. Edward (Ed) Borasky