numerical computing environments

[offroad]

Hi there,

Does anyone of you work with numerical computing environments such MatLab and Scilab?

If so, tell a bit about your work.

I work with both. I like Matlab, but I don't have enough license files to install it on all machines. Scilab is great too.

However, I downloaded the newest version of Scilab, 5.2.x and found it a bit 'slow', and couldn't understand why its graphic windows can't be maximized without messing the graph.

Thanks for your time.

offroad

[lmh]

If cash is an issue, have you considered "R"? It's free, very flexible, and has a huge range of ready-written packages.

http://www.stats.bri.ac.uk/R/

R is also gaining ground in the metabolomics community because of the ready-written (and ever improving) package xcms, also free:

http://www.bioconductor.org/

The down-side of R is that it assumes the user has brain-cells and an inclination to use them in bioinformatics; it also expects that the user should be able to type, and differentiate between big and little letters, which can be challenging to those of us who have grown used to Windows-based applications with big friendly green buttons.

Most help-files associated with R and its packages were clearly written by a sadistic computer, or (worse) a bioinformaticist/computational expert.

Apart from that, it's great.

[offroad]

Yes, I used R when I was a linux user, just a couple of years ago. Slackware linux, miss it so badly.

But...then... They introduced me to Statistica, very powerful tool. It makes R sound like... children's play. But the problem is... when we start palying with huge amount of data, Statistica takes ages to process and finish a calculation. Then we must build our own application on platforms like FORTRAN, Scilab or Matlab.

If you are wondering about how big Matlab is, keep in mind that it will take about 5 GB of your hard disk. Of course, it ain't any cheap.

[HW Mueller]

Imh, your links don´t work here (Germany).

[offroad]

I believe this will do:

http://www.r-project.org/

[HW Mueller]

That got it there.

[offroad]

Are you going to do some calculation? what is it about?

[offroad]

However, I downloaded the newest version of Scilab, 5.2.x and found it a bit 'slow', and couldn't understand why its graphic windows can't be maximized without messing the graph.

I've just found out, it's some Vista issue. It works perfectly on XP.

[HW Mueller]

I am still helping some students, which means I have to be "out front" on statistics, especially to stay clear of things that I cannot handle mathematically.
I was interested in this chain, because I wondered what people do with these math programs. Always wanted to get one, but couldn´t convince med. people that we need it. I simply wanted to do some algebra, and calculus for pH, equilibrium, curve fitting.
Thus, I have the question: Why would someone use these for doing statistics with all the statistic programs around?

[offroad]

Sadly some people don't like it, they prefer do it by hand, especially if there's not such amount of data to be processed.

You guys feel free to post some questions here. Since I don't have much experience to help you in chromatography yet, I'll try to help by giving some tips on how to solve problems numerically.

[lmh]

HW Mueller,

I wouldn't use R for simple calculations. I use it because xcms is proving a very powerful general peak-finder for LC-MS data, and is gaining wide acceptance in the metabolomic world. R can do PCA in one typed line, and PLS is available in several forms through the many extra packages written for R.

R (probably like any of these packages) does have a few useful aspects for me, for instance given a biggish matrix of peaks-found (rows) against samples analysed (columns), and a further matrix of qc sample run multiple times, it is very easy to find the relative standard deviation of each peak in the qc runs, and extract a subset of results from the other matrix containing only those rows whose rel. s.d. in qc samples is better than a certain value (i.e. forget the peaks that have been measured unreliably).

Yes, Matlab can do everything that R can (and vice versa!), and Matlab has a much nicer interface, but I can't justify the costs associated with Matlab.

Feelings seem to run very high amongst informaticists about which package is the best. The more sane informaticists (my personal assessment!) seem to feel it's not such an issue as all the packages do much the same job and offer similar facilities.

Thanks for letting me know those links don't work everywhere. I assumed they ought to. I'll offer the alternative too, next time... It is very hard to Google search for "R".

[offroad]

lmh,

I know Matlab is not any cheap at al, but its cousin/brother, Scilab, is free.

If you catch yourself bored on a rainy weekend, maybe you want to try scilab.

I might say Scilab is twice as fast as R, simpler, can process lots of data etc, but it's up to you to give it a go/shot.

Regards

Offroad

[lmh]

Thanks for the hints! To be honest, I can't really compare the packages greatly on speed as all my applications are so small that my typing is the bottle-neck, not the computational time. I'm only using R because of xcms and wouldn't claim to be an expert.