Package overview

From Madagascar
Jump to navigation Jump to search
Fotolia 5127720 XS.jpg

The mission of the Madagascar project is to provide a shared research environment for computational data analysis in geophysics and related fields.

The Madagascar environment consists of:

  1. Standalone programs for out-of-core numerical analysis;
  2. Standalone programs for geophysical data processing and imaging;
  3. A development kit for C, C++, Fortran-77, Fortran-90, Python, Matlab, and Octave;
  4. A framework for reproducible numerical experiments, based on SCons;
  5. A framework for scientific publications, based on SCons and LaTeX;
  6. A collection of reproducible scientific articles also used as usage examples and regression tests for the standalone programs;
  7. A collection of datasets used as input to reproducible numerical experiments.

This guide serves as a brief introduction of different components and shows how they all fit together.

How to find your way around Madagascar

If you are new to Madagascar, start by looking at the list of reproducible papers. If any of these papers looks close to your interests, follow the links until you find a figure with a "wrench" button under it Configure.png. Click on the wrench, and it will open a computational recipe used for generating the figure (the SConstruct file). You can copy this recipe to your computer or simply find it already existing in the "book" subtree under the Madagascar source directory. For example, the recipe at http://www.reproducibility.org/RSF/book/rsf/rsf/sfmath.html also exists in the file RSFSRC/book/rsf/rsf/sfmath/SConstruct. After copying or location the appropriate SConstruct file, run

scons view

on the command line to generate all the figures in the selected project and to display them on your screen. For example, try

bash$ cd RSFSRC/book/rsf/rsf/sfmath
bash$ scons view

where bash$ stands for the Unix prompt and RSFSRC stands for the Madagascar source directory. The output should look like

scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
/RSFROOT/bin/sfmath n1=629 d1=0.01 o1=0 n2=40 d2=1 o2=5 output="x2*(8+sin(6*x1+x2/10))" | /RSFROOT/bin/sfrtoc | 
/RSFROOT/bin/sfmath output="input*exp(I*x1)" | /RSFROOT/bin/sfgraph title=Rose screenratio=1 wantaxis=n > Fig/rose.vpl
/home/fomels/RSFROOT/bin/xtpen Fig/rose.vpl
scons: done building targets.

with a picture appearing on your screen.

If there are several figures in the recipe, you can run

scons figurename.view

to display individual figures. To remove all the files that scons view generated, run

scons -c view

If you want to know what commands will be executed to generate the figures before actually running them, try

scons -n view

You can output this command to a file

scons -n -Q view > script.sh

and use script.sh as a shell script. If you are to make modifications to the data processing recipe defined in SConstruct (changing parameters or trying new data), executing scons is more powerful and convenient then running shell scripts.

[TO BE CONTINUED]

Standalone programs

Most programs act as filters on input data and can be chained through Unix pipes, i.e.: <bash> < data.rsf sfwindow n1=100 | sfbandpass fhi=60 > data2.rsf </bash>

This approach follows the Unix philosophy, as formulated by Doug McIlroy, the inventor of Unix pipes (Salus, 1994[1]):

  1. Write programs that do one thing and do it well.
  2. Write programs to work together.
  3. Write programs to handle text streams, because that is a universal interface.

Documentation has designed in a layered fashion. Following the Unix convention, programs have brief man pages, which explain the program purpose and parameters. These pages can also be accessed by running a program without any arguments, or online. A more detailed level of documentation, with basic equations, figures and examples, is provided through the wiki in the [Guide to madagascar programs | Guide to programs]. The Task-centric program list categorizes and briefly describes all components. At an even higher level of details, the programs can be seen in actual use in the [Reproducible Documents]. Search capabilities are currently provided by the sfdoc utility.

Madagascar data formats

For data, Madagascar uses the Regularly Sampled Format (RSF), which is based on the concept of hypercubes (n-D arrays, or regularly sampled functions of several variables), much like the SEPlib (its closest relative), DDS, or the regularly-sampled version of the Javaseis format (SVF). Up to 9 dimensions are supported. For 1D it is conceptually analogous to a time series, for 2D to a raster image, and for 3D to a voxel volume. The format (actually a metaformat) makes use of a ASCII file with metadata (information about the data), including a pointer (in= parameter) to the location of the file with the actual data values. Irregularly sampled data are currently handled as a pair of datasets, one containing data and the second containing the corresponding irregular geometry information. Programs for conversion to and from other formats such as SEG-Y and SU are provided.

For graphics, Madagascar currently uses the Vplot vector graphics format. Converters to other graphics formats (Postscript, PNG, GIF, JPEG) are also provided.

Reproducible documents

A reproducible document consists of LaTeX source combined with SCons rules required to fully build the documents. These rules are expressed in terms of SCons extensions that are provided as part of Madagascar.

This is the key to the reproducibility aspect of Madagascar. An introduction to reproducible Madagascar documents is at Reproducible_computational_experiments_using_SCons .

Madagascar Display: Vplot

In contrast to most other Madagascar Components, graphics components produce Vplot data as output.

Vplot is a device independent graphics format that allows both vector and raster elements (as such, it is comparable to Postscript). Vplot files are interpreted by a number of output devices. Its typical usage is for a visual display in X-windows. A list of them is provided on the wiki.

Here is an example of a Madagascar pipe. In this case it takes a subsection of a file, low-pass filters it, and saves the result

<bash> < data.rsf sfwindow n1=100 | sfbandpass fhi=60 > data2.rsf </bash>

In this more elaborate case, the final output is passed to a graphics program and plotted.

<bash> < data.rsf sfwindow n1=100 | sfbandpass fhi=60 | sfcontour | xtpen </bash>

More extensive examples are seen at Guide to madagascar programs . The novice reader should probably read the material below before proceeding to that page.

Reproducibility and Project Management

Madagascar uses and extends SCons, an open-source software construction package, to document and maintain data processing flows. Documented projects become computational recipes that can be easily exchanged among Madagascar users.

SCons is a rule-based package in Python typically used as a build system analogous to make. Familiarity with any build system will be helpful in understanding SCons. SCons statements, as python statements, are invoked in the sequence they are written, but as such they only define rules. The rules are invoked in accordance with a dependency graph which SCons builds based on those rules. Components regarded as "up-to-date" are not rebuilt.

SCons allows user-contributed Builders (meta-rule categories) and Madagascar uses this capability extensively. The idea is that building an output file based on a workflow chain is very much analogous to building a software package based on a software tool chain. The calculation is seen simply as a build with dependencies. This is a considerable benefit in developing alternative workflows using a given dataset. The system maintains an awareness of already completed calculations. Without user intervention, redundant calculations are avoided.

Madagascar calculations are thus expressed as SCons scripts (SConstruct files). SCons extensions follow SCons conventions in beginning with an uppercase letter. The most common Madagascar extensions are Flow(), Result(), and End(). A Flow() invocation wraps Madagascar computational components. Result() is a version of Flow() with a graphical output. Finally an End() actually invokes the default rules for multiple results.

Finally, Madagascar enables a collection of reproducible documents, organized in living books. Each reproducible book contains a collection of Madagascar recipes (SConstruct files) used to generate book figures. The recipes cover a variety of data processing and imaging tasks described in the books. Figures and recipes serve dual purpose with respect to Madagascar maintenance. They provide demos for introducing new users to the functionality of the package and, at the same time, regression tests for assuring the system stability under change.

How it All Comes Together

Here is an example code, described in detail on the SCons page.

<python> from rsfproj import *

  1. Download the input data file

Fetch('lena.img','imgs')

  1. Create RSF header

Flow('lena.hdr','lena.img',

    'echo n1=512 n2=513 in=$SOURCE data_format=native_uchar',
    stdin=0)

  1. Convert to floating point and window out first trace

Flow('lena','lena.hdr','dd type=float | window f2=1')

  1. Display

Result('lena',

      
      sfgrey title="Hello, World!" transp=n color=b bias=128
      clip=100 screenratio=1 
      )

  1. Wrap up

End() </python>

Getting Madagascar

Madagascar runs on Unix/Linux platforms, including MacOS X and Unix emulations under Miscrosoft Windows. Its installation requires, at a minimum, a working C compiler and Python. Most users will also want an X-Window system on their desktop. See download and installation instructions.

License

The Madagascar package is released in an open-source form under the standard GNU GPL license. In simple words, there are no restrictions on the use of the software (including copying, modifying, selling, etc.) However, there are restrictions on the software redistribution intended to prevent the package from losing its open-source status. Users are encourages to submit their modifications back to the original distribution to the benefit of the whole user community.

Why the Name "Madagascar"?

Whimsy, really. It seems easier to remember than the previous name "RSF", and it provides us lots of interesting mascots.

Madagascar Community

Madagascar seeks to become an open and active open source community. Active mailing lists are maintained and annual meetings take place. See

Your participation is welcome.

History

In the present form, the Madagascar package, while being completely written from scratch, borrows ideas from the design of SEPlib, a publicly available software package, maintained by Bob Clapp at the Stanford Exploration Project. Generations of SEP students and researchers contributed to SEPlib. Most important contributions came from Rob Clayton, Jon Claerbout, Dave Hale, Stew Levin, Rick Ottolini, Joe Dellinger, Steve Cole, Dave Nichols, Martin Karrenbach, Biondo Biondi, and Bob Clapp.

Madagascar also borrows ideas from Seismic Unix (SU), a package maintained by John Stockwell at the Center for Wave Phenomenon at the Colorado School of Mines (Stockwell, 1997[2];Stockwell, 1999[3]). Main contributors to SU included Einar Kjartansson, Shuki Ronen, Jack Cohen, Chris Liner, Dave Hale, and John Stockwell. SU is open-source software (distributed with BSD-style license) starting with release 40 (April 10, 2007).

References

  1. Salus, P. H., 1994, A quarter-century of Unix: Addison-Wesley.
  2. Stockwell, J. W., 1997, Free software in education: A case study of CWP/SU: Seismic Unix: The Leading Edge, 16, 1045--1049.
  3. --------, 1999, The CWP/SU: Seismic Un*x package: Computers and Geosciences, 25, 415--419.