Does anyone actively use SCILAB or OCTAVE for high performance data analysis? Can anyone share their experiences with SCILAB, OCTAVE, or Matlab?
I'm seriously considering a replacement for Matlab in my academic environment in order to drop the licensing cost.
It is perfectly possible to use GNU Octave within the context of computationally-intensive tasks (e.g. see an interesting comparison: http://mfkp.org/INRMM/article/11523561 ). For example, it is possible to use it for large scale environmental modelling. This also means that data-parallelism could be exploited with a disciplined programming style (i.e. by taking seriously array programming so that the low-level numerical libraries on which GNU Octave relies are efficiently used).
GNU Octave has the advantage, over other free software tools, of providing very high compatibility with the MATLAB language. Actually, the Octave language (http://www.ohloh.net/languages/octave ) and the MATLAB language (http://www.ohloh.net/languages/matlab ) may be considered as dialects with differences which generally do not prevent scripts and libraries to run in both GNU Octave and MATLAB env. (see also http://wiki.octave.org/FAQ#Porting_programs_from_Matlab_to_Octave and a cross-language library for numerical modelling with potentially large arrays of data http://mastrave.org/doc/mtv_m Chapter Semantic Array Programming for environmental modelling: appl...
)Among many examples of computationally intense applications, you could have a look to Conference Paper Architecture of a Pan-European Framework for Integrated Soil...
which describes the use of semantic array programming with GNU Octave and the Mastrave modelling library for supporting a continental-scale modelling architecture for assessing soil erosion by water.Three minor points. First, a terse array programming style is of course and more generally recommendable per se when dealing with array-oriented languages such as MATLAB, irrespective of whether the computing environment is GNU Octave or MATLAB – or other compatible interpreters.
Second, if the model to implement is intended to be reused as a serious stuff (or just if it is large enough for you not to be able to keep in mind all details *together*), then the additional overhead due to the systematic use of semantic checks (Conference Paper Software uncertainty in integrated environmental modelling: ...
Conference Paper Toward open science at the European scale: Geospatial Semant...
Chapter An architecture for adaptive robust modelling of wildfire be...
Chapter Continental-scale living forest biomass and carbon stock: a ...
Chapter The European Atlas of Forest Tree Species: modelling, data a...
Book Robust modelling of the impacts of climate change on the hab...
) is generally more than compensated by the boost in understandability and maintainability. 100% agree with H. Sutter on exception safety which is never "just an implementation detail" (http://mfkp.org/INRMM/article/11861232 ).Third, GNU Octave is an advanced free software package and is suitable to be configured for improving its response to specific needs (see http://www.gnu.org/software/octave/doc/interpreter/Running-Configure-and-Make.html ). Personally, I have to say I collected several evidences supporting the opinion of D. Knuth on premature optimisation.
Concerning the specific topic of high performance computing and array programming based also on GNU Octave, one might find of some potential use Article Architecture of Environmental Risk Modelling: for a faster a...
Enginsoft Spa in Padova does it. Contact SIlvia Poles: [email protected]
To be more precise, Enginsoft uses Scilab. See also http://www.openeering.com/
In Short, No. Octave and Scilab are great for testing an algorithm for correctness and for fast code development. However, these environments have a lot of built-in performance overhead which prevent optimization of execution time. They are, in fact, higher level languages which allow ease of understanding to the developer and portability of platforms. Lower level languages like C and Fortran allow you the flexibility to optimize not only at the code level, but at the compiler level as well. However, the cost of optimization is that code must be fine tuned at all levels for the specific target architecture (computer). Such flexibility is unavailable in current versions of octave and scilab for large scale high performance (multi-node, multi-core) computing
It is perfectly possible to use GNU Octave within the context of computationally-intensive tasks (e.g. see an interesting comparison: http://mfkp.org/INRMM/article/11523561 ). For example, it is possible to use it for large scale environmental modelling. This also means that data-parallelism could be exploited with a disciplined programming style (i.e. by taking seriously array programming so that the low-level numerical libraries on which GNU Octave relies are efficiently used).
GNU Octave has the advantage, over other free software tools, of providing very high compatibility with the MATLAB language. Actually, the Octave language (http://www.ohloh.net/languages/octave ) and the MATLAB language (http://www.ohloh.net/languages/matlab ) may be considered as dialects with differences which generally do not prevent scripts and libraries to run in both GNU Octave and MATLAB env. (see also http://wiki.octave.org/FAQ#Porting_programs_from_Matlab_to_Octave and a cross-language library for numerical modelling with potentially large arrays of data http://mastrave.org/doc/mtv_m Chapter Semantic Array Programming for environmental modelling: appl...
)Among many examples of computationally intense applications, you could have a look to Conference Paper Architecture of a Pan-European Framework for Integrated Soil...
which describes the use of semantic array programming with GNU Octave and the Mastrave modelling library for supporting a continental-scale modelling architecture for assessing soil erosion by water.Three minor points. First, a terse array programming style is of course and more generally recommendable per se when dealing with array-oriented languages such as MATLAB, irrespective of whether the computing environment is GNU Octave or MATLAB – or other compatible interpreters.
Second, if the model to implement is intended to be reused as a serious stuff (or just if it is large enough for you not to be able to keep in mind all details *together*), then the additional overhead due to the systematic use of semantic checks (Conference Paper Software uncertainty in integrated environmental modelling: ...
Conference Paper Toward open science at the European scale: Geospatial Semant...
Chapter An architecture for adaptive robust modelling of wildfire be...
Chapter Continental-scale living forest biomass and carbon stock: a ...
Chapter The European Atlas of Forest Tree Species: modelling, data a...
Book Robust modelling of the impacts of climate change on the hab...
) is generally more than compensated by the boost in understandability and maintainability. 100% agree with H. Sutter on exception safety which is never "just an implementation detail" (http://mfkp.org/INRMM/article/11861232 ).Third, GNU Octave is an advanced free software package and is suitable to be configured for improving its response to specific needs (see http://www.gnu.org/software/octave/doc/interpreter/Running-Configure-and-Make.html ). Personally, I have to say I collected several evidences supporting the opinion of D. Knuth on premature optimisation.
Concerning the specific topic of high performance computing and array programming based also on GNU Octave, one might find of some potential use Article Architecture of Environmental Risk Modelling: for a faster a...
Well I used it all the time. In terms of performance better use SciPy.
Thanks Paul Delgado
Indeed i agree about your comparison between compiler vs script languages.
To be a bit more focused I was interested in a matlab replacement of some sort.
There are many high-performance tasks perfectly duable in matlab,
for example there is very little overhead in performance in spectral analysis as FFT calls are actually compiled. In essence in matlab you don't code as in C, but you try to describe everything in macro SIMD instructions, and there are many problem that are quite fit .
Another example is eigenvalue extraction. It is very challenging to have a code faster then matlab native call. Even in Fortran or C i would simply call some compiled LAPACK or BLAS function to do it, that is basically what matlab does. It is true that in C or Fortran there is indeed an advantage because you can create more hardware specific compiled version of your code, but the development load is not comparable.
Indeed there are many codes that does not fit macro instruction description, e.g. that requires a lot of iterations (e.g. finite difference time domain or finite element solver) that are very slow in matlab.
Thanks Melissa for the tip on Python/SciPy i will check it out.
Marco
Personally, I don't use it because I have Matlab licence, and (most importantly) because I'm rather skilled in Matlab, and now I have no time to change. I used Scilab some years ago (2004) due to licence problems, and at that time I thought that it can replace Matlab. I cannot state now because I don't follow Scilab any more, but I think that it's definitely a very good software for you.
I use Octave in my Scientific Programming course. There tends to be a small hiccup here or there with various Octave Forge releases for Windows (rarely do more than 1/3rd of my students choose to run under linux), but nothing that can't be dealt with fairly easily. I'm planning to use SCILAB XCOS (a very good Simulink "clone" IMHO) in a Spring course in order to model some feedback control systems.
I primarily use MATLAB, but do occasionally pull up Octave to make sure the code I write is compatible. The primary reason I still pay for a MATLAB license is that Octave's development environment is still pretty weak relative to MATLAB's. The extra time it takes to debug in Octave isn't worth it for me. As a secondary effect, I do occasionally want to generate compilable code quickly from my MATLAB scripts, and Mathworks's autocoding tools make it easy for me to be lazy.
I have used Matlab during my studies, and now I use SciLab. SciLab has a Matlab's function translator, which can be helpful if you want to re-use your m. files. Also, last version of Scilab has an appearance similar to Matlab.
Well already you get the enough information that can give you the exact thing that you want. I want to add few more things but if you are restricted with Octave and Scilab then it will not be fruitful.
It my general experience that I am having on Data Analysis, I feel R software will be good to deal with data analysis more efficiently.
If you thing that you can do enough time explore the best one then go through that....
I have license of MATLAB but still I teach algorithms using OCTAVE. I like to use GNU software and I am a promoter of GNU softwares for scientific computing. This is due to the fact that many of my students are not having licenses of commercial softwares.
Yes, I do.
In the company I work for we use Scilab to computational calculations. For huge data amounts, Scilab is commited to external in-house fortran/C routines, and the performance is super!
I'm currently using python (NumPy, Scipy) Octave and Scilab depending, basically on the libraries I need for the data analysis.
I used in my courses (University of Bari, Italy), Octave, Scilab and R, but now I'm using Python, with numpy and scipy. Pythonxy in Windows and spider for Linux are good ide for scientific computing. I developed some Python interface for my FORTRAN 90 codes and was easier than in matlab. I still use Matlab for research, but I think Python is a good alternative, my students like it ....
Yes, there exist one more reason to prefer GNU Octave: it is more accurate.
Try to insert the next sequence of exact binary numbers in both Matlab and Octave:
*****************************************************************************************************
octave:1> format rat
octave:2> i=[0:1:29];
octave:3> y=100+(96+2.*i)/(2^10)
y =
Columns 1 through 10:
3203/32 51249/512 25625/256 51251/512 12813/128 51253/512 25627/256 51255/512 6407/64 51257/512
Columns 11 through 20:
25629/256 51259/512 12815/128 51261/512 25631/256 51263/512 801/8 51265/512 25633/256 51267/512
Columns 21 through 30:
12817/128 51269/512 25635/256 51271/512 6409/64 51273/512 25637/256 51275/512 12819/128 51277/512
*****************************************************************************************************
>> format rat
>> i=[0:1:29];
>> y=100+(96+2.*i)/(2^10)
y =
Columns 1 through 4
3203/32 9409/94 4104/41 26126/261
Columns 5 through 8
12813/128 2903/29 25627/256 12113/121
Columns 9 through 12
6407/64 51257/512 5306/53 26931/269
Columns 13 through 16
12815/128 4205/42 25631/256 6508/65
Columns 17 through 20
801/8 6308/63 25633/256 10714/107
Columns 21 through 24
12817/128 14119/141 11716/117 13719/137
Columns 25 through 28
6409/64 51273/512 8312/83 15723/157
Columns 29 through 30
12819/128 13320/133
Look at Table 2 of page 15 from the attached Ebook (it is in greek, but tables are in english) and compare the absolute accuracy of Octave with the errors of Matlab.
You will see that the very expensive Matlab cannot represent even binary numbers with accuracy!
Book Σημειώσεις Εργαστηρίου Γραμμικών Μαθηματικών
Dear Krzysztof
thanks, i have a class of 71 students and I would like to use a computational platform that does not require a very expensive licensing.
Marco
Haven't seen it for years, but there used to be a library called Blitz++ which included nice matrix operations like that of Matlab. The beauty of using this is that you can then just pull in CGS, ICCG and other matrix inversion methods very quickly from numerical recipes in C++. Obviously its not as easy to programme as numpy or scipy but you get a lot of extra libraries for free this way.
I use Octave regularly. Its advantage us platform independence. I develop programs on Linux and some colleagues use them on Windows. I even use it for "near real time" data analysis. The measuring instrument is about 100 km apart. I get data once per hour via internet. I do some reformatting and processing by methods of mathematical gnostics. The results are stored in a shared repository so that colleagues can access them and selected results are plotted and made online on the web site. Selected results are automatically trasmitted to a central databes located 1500 km apart. The system runs on Linux. Algorithms of mathematical gostics were programmed by me in Octave. Transfer and program invocation is written in a set of bash and perl scripts, plots are generated by gnuplot and the web site is built upon Apache with PHP. The PNG files generated by gnuplot are small so the fastest way is to store them as sttic images in a directory tree and the web pages are generated by a simple PHP script.
I am generally a command-line-man, I do not mind if I have to type commands and Octave on all platforms supports completion, it is easy to use if you learn a few tricks. It might be difficult to prepare big matrices but using a text editor you can make a file containing
a = [
1 2 3
4 5 6];
and then read the file by the source command. You can have a matrix in a spreadsheet and convert it to such a file quite quickly. If your students have to convert large matrices repeatedly, they will hopefully learn to think algorithmically and they may understand, that a spreadsheet s not always the best tool because data can be manipulated faster and more easily by awk or perl or other free scripting programs.
I have been using GNU Octave for fifteen years, and it is great! (and all my code runs both in GNUOctave and Matlab)
I am using GNU Octave ( http://octave.org ) in the context of model order reduction of large-scale nonlinear control systems. From my point of view Octave is the best replacement of Matlab. Octave is open source and replicates the Matlab language and syntax best and the differences between Octave and Matlab are documented very well ( http://wiki.octave.org ). Scilab has some syntactic differences and Freemat has not implemented the same amount of commands. The plot backend is GNU Plot which allows very comprehensive plots like Matlab, not only to a window but also to eps, pdf, png and even to the terminal as ASCII plot. Furthermore Octave provides packages ( http://octave.sourceforge.net ) which correspond in part to the Matlab toolboxes. By default Octave does not provide a graphical user interface, just a plain terminal, which I think is an advantage; yet if required, frontends like QtOctave can be installed. In many cases Octave is not as fast as Matlab, but based on the same libraries, and very few Matlab commands are not available yet. Both these issues decrease in every new version of Octave. (Effective: october 2013)
Dear Marco,
Since several years ago, we use the scilab for numerical computations in place of Matlab due for the same reason given by you. One of the most amazing features of the Scilab for high requirement computing is that you can compile the source code and optimize for your plataform. This is a characteristic that Matlab hasn't, and is really very adventageous in our work.
Else Scilab count with a flexible interface for GUI applications of appreciate quality.
We can save a Matlab licence for use the toolboxes that are very developed by Mathworks and none of the gnu software has yet.
You might also like to take a look at MLAB (at www.civilized.com)
MLAB is a commercial product (born at NIH) but without recurring fees.
It's main distinguishing feature is it's ability to fit ODE-defined models
such as arise in chemical kinetics, physiology and pharmacology.
(The attachment is an example of MLAB.)
I have been struggling with Scilab for about one year, for simple data computation and plots, migrating my own scripts from Matlab. My answer is: don't use Scilab. I spent a tremendous amount of time trying to figure out what would be the next bug. In particular, using libraries of scripts, getting decent plots on different computers, and formatting numbers, have been among my nightmares. Documentation and forums are either poor or just not helping. I am now considering moving to Octave, or C++.
I am scilab and C. Scilab is easy and xcos available in is useful GUI.
- Badges
- Science topic
- Similar topics
- Computer Science
- Computing