Your experiences with Matlab/F#/R for data analysis and modeling algorithms

I've been using F# for a while now to model algorithms before coding them in C++, and also using it afterwards to check the results of the C++ code, and also against real-world recorded data.

For the modeling side of things, it's very handy, but for the 'data mashup' kind of stuff, pulling in data from CSV and other sources, generating statistics, drawing charts etc., my colleague teases me no end ("why are you coding that yourself? It's built in to MatLab").

And I have another colleague who swears by R, which also has charting stuff 'built-in'.

I know that MatLab, R and F# are not strictly comparable, so I'm not asking for a 'feature comparison shoot out'. I just wondered what other people are using for these kind of pre- and post-analysis scenarios, and how happy they are with it.开发者_JS百科

(If there's anyone out there working on wrapping Microsoft Charts into something F#-friendly, let me know, I'd be happy to participate...)

(Note: answers to this question will be subjective, but based on experience, please)

I have very little experience with F#, but regarding C++/Matlab/R: If the speed of your program's execution is the most important, use C++. If speed of implementation is the most important, use Matlab or R. This is true for a number of reasons, not the least of which is their massive libraries of math/stats packages.

Both Matlab and R can be sped up through parallelism: so generally, I think that speed and quality of implementation should be a bigger concern. That's where the real "value" of programming is taking place, in the design of the application. It's not a minor proposition if you can write 3 or 4 good R programs in the same time it takes you to write 1 good C++ program.

Regarding F#: so far as it is part of Microsoft's framework, it must have a lot to offer. If you're developing in Visual Studio or working on a big .Net project (for instance), it might make sense to use F#. On the other hand, you can call both Matlab and R from .Net applications, so I would probably argue that their libraries should be a bigger concern. For instance, see this article as an example for R and the Matlab Builder.

Long story short: comparing F# and Matlab/R isn't a good comparison. F# is a general purpose programming language, while Matlab/R can be viewed as massive mathematical/data analysis toolkits. Some people call Matlab or R from F# in order to take advantage of each language's benefits (e.g. see this discussion, this article on Matlab/F#, or this article on R/F#).

So far as charting is concerned: R is extremely strong on this front. Have a look at the graphics view on CRAN and this series of posts on the LearnR blog about Lattice and ggplot2.

I've worked a bit with matlab and python/pylab for these purposes. What these tools have 'built-in' is a programming environment, a shell, and gui tools designed for quickly looking at data from a variety of sources.

In a few commands, you can go from having a csv file to interactive plots on the screen, then to an image export in just about any format. It takes a minute or two to go from data to visualization once you have the hang of it. I would imagine this is uncommon in the C++ world (although I have seen some professors with pretty impressive work-flows).

I've tried R, but I can't say much useful about it. It seems to offer about the same set of features, but it may be troublesome to Google for support.

If you are spending more than a couple minutes getting from data to plot using your current method, it's definitely worth learning one of these environments. The best choice depends on your colleagues, your work environment, experience, and your budget.

This is a reasonable close double to the previous question on suitable functional language for scientific/statistical computing so you may want to peruse the long and detailed answers there.

Answers depends, as so often, on your experience and prior language training. I very much prefer R for data munging / modeling / visualization.

I use R because on the one hand it has everything built in and on the other hand you can still manipulate almost everything or start from scratch. Nevertheless, R is rather slow for heavy calculations (although I do all my Monte Carlo simulations in it).

I would say that Matlab is best for the availability of mathematical functionalities in general, R is best for data input/manipulation/visualisation/analysis/etc., and C++ for high-speed subroutines. You can by the way easily integrate C++ (or C, fortran, ...) code in R. Why not read and manipulate input data in R, apply the models in C++, and analyse/visualize output back in R?

I always prototype my models in MATLAB. If my prototype is fast enough, I refactor and it's done. If not, I go back and implement certain functions in C to be called by MATLAB. This requires knowledge of a low level language, which I think is always going to be the case if you are doing anything that is technically challenging.

I'm intrigued with this Lisp flavor if it ever gets off the ground.