It's powerful interface is really useful. And for academics it's not really expensive.

I worked a lot with octave and Python in the past. Octave is good for most of the simple tasks, and Python is my first choice for anything bigger. But the integration of the Editor, Debugger, easy GUI creation and load- and save-tools of MATLAB are so well made, it almost makes you forget the language is not open source.

However, I think there are two major drawbacks of MATLAB: First, it is very possible that your Code won't work for someone else if it's using some functions of a specific toolbox that the other one doesn't own. The other problem is the high prize outside of academia. Of course thats on purpose. When you've learned MATLAB for years you'll want to use it for your job outside academia, too. But for a small company the high prize can be a problem. In my opinion, this increases the problem of the many toolkit-functions.

The conclusion I drew for myself is to use MATLAB as developement language. It's easy to code and debug. If you stay in academics or have the money for MATLAB, everything's fine. But if you want to work with many people, and these may have no access to MATLAB, then it's better to use an open-source language like Python.

What do you think?

Much of this question is irrelevant - The MathWorks offers substantial discounts for academic use of Matlab, to both students, professors, and academic researchers. I think also much of the existing toolboxes are extensions developed originally in academia, showing that at least in some cases there is a collaborative relationship between the MathWorks and researchers.

It is very difficult to match the expertise that has gone into Matlab, for anyone that has dug into the numerical techniques, automatic detection of ill-conditioned or other special cases, and automatic selection of the appropriate algorithms, the amount of machinery that "just works" is amazing. This goes way beyond the "matrix library" level.

It's powerful interface is really useful. And for academics it's not really expensive.

I worked a lot with octave and Python in the past. Octave is good for most of the simple tasks, and Python is my first choice for anything bigger. But the integration of the Editor, Debugger, easy GUI creation and load- and save-tools of MATLAB are so well made, it almost makes you forget the language is not open source.

However, I think there are two major drawbacks of MATLAB: First, it is very possible that your Code won't work for someone else if it's using some functions of a specific toolbox that the other one doesn't own. The other problem is the high prize outside of academia. Of course thats on purpose. When you've learned MATLAB for years you'll want to use it for your job outside academia, too. But for a small company the high prize can be a problem. In my opinion, this increases the problem of the many toolkit-functions.

The conclusion I drew for myself is to use MATLAB as developement language. It's easy to code and debug. If you stay in academics or have the money for MATLAB, everything's fine. But if you want to work with many people, and these may have no access to MATLAB, then it's better to use an open-source language like Python.

What do you think?

In transitioning to the commercial world the price might be startling, or a shock, or even a challenge to justify - then again, compared to the price of CAD systems, a commercial RTOS or even the cost of Visual Studio, MSDN and Office tools - other software tools necessary to particular businesses - Matlab, even including toolboxes is a comparable business expense. A deductible capital equipment expense.

Python, C++, Perl with XXX numerical library *is not* a replacement for Matlab, different tools, different applications and different users.

I'm not sure I'd agree with you there James Taylor. Everything you've indicated in MATLAB is easy replaced by faster, better alternatives in Python. I think the comparison of MATLAB to CAD systems is a fallacious comparison.

I would agree that neither C++ nor Perl are replacements for MATLAB, but let's look at Python and Octave, since those are the only two pieces of software that I consider to have similar audiences.

Performance-wise Python's numpy library is considerably faster than MATLAB by a factor of 2-3 in most computational tasks (https://modelingguru.nasa.gov/docs/DOC-1762) Add to that, python includes a library, (gnumpy) that allows you to run code on your gpu, and the performance gains easily increase to an order of magnitude above that. Octave on the other hand is significantly slower than MATLAB (by about the same factor)

Utility-wise MATLAB includes more signal processing and non-linear optimization toolkits than Python, but Python absolutely blows MATLAB out of the water when it comes to libraries for anything else. There are a few packages aimed at replacing MATLAB (such as matplotlib) that end up going above and beyond the capabilities of MATLAB, and then there are large swaths of functionality absent, or poorly implemented in MATLAB and Octave that are trivial in Python (such as anything involving the internet)

As a part of that we can consider looking at cross-platform operability and packaging. Python includes pypi, which makes the additional of any package, including any dependency managing trivial across systems, this is opposed to MATLAB, where dependency management is proprietary, and as referenced above, extremely brittle. Octave's dependency management is practically non-existent, though it copies MATLAB in many regards on this front.

Pricing-wise, you've raised the point that MATLAB is "not expensive" for academics, but a hundred dollar license is absurd for an academic in a third world country. Within the US, Canada, or EU it might be a trivial expense, but this is absolutely not the case for people outside of that small bubble.

In a very related way let's talk about users. Currently MATLAB is used exclusively by Academia. Outside of academia, in professional development circles MATLAB is generall loathed and looked down upon, while most languages outside of MATLAB and FORTRAN enjoy a similar fate for those within academia.

The problem here is that we have two massive communities that are unable to benefit from each other's expertise. The professional development community within Python has recreated a staggering amount of the work done within academia in the past decade, yet this performant development manages to stay away from the hands of people that really need it because Python is largely eschewed in academia.

Looking from an economic justice standpoint, those that dearly need to benefit from the advantages of academic research cannot benefit from any advances made in academia until someone with free time and access to an academic network puts the same code into Python. If this is a transition that has to happen anyway why not simply move over to another development environment that is already significantly more performant and widely adopted at large than MATLAB?

Look into the actual implementation of the basic functionality. The academic population you're referring to is as often numerically and computationally challenged. You are not likely to be in the position to add missing functionality, creating a libraries and assembling algorithms to address the variety of problems inherent in any non-trival numerical modeling or development problem.

The problem with picking a comparable benchmark is that you have to pick benchmarks where there is comparable functionality - not problems where the algorithm doesn't exist within Python libraries, or where the Python library gives the wrong answer.

In any non-applied math, non-computer science discipline, you are likely to have been introduced to the basic principles of the mathematics for your field, but never in my experience will you be prepared to provide professional competence in the implementation and the required diligence in V&V to assure that your methods and code are valid and robustly implemented.

Reading "Numerical Recipes" doesn't prepare anyone for the challenge of creating a *good* numerical library; too many think that is even an adequate starting point. If you don't know the number of ways a large matrix inversion can go wrong, how well prepared are you to implement a Yule-Walker based spectral estimate or visualization code, or a thousand other complicated and disparate tools available in Matlab and it's toolboxes.

As far as used "exclusively by Academia"; Matlab has been used at every corporation I've worked with or for, because it is one of the few places one can model large filtering and algorithmic systems, and include good noise and quantization models. Philips Medical and Medtronic use it extensively to model and specify diagnostic systems. Boeing uses it to model aerospace systems. Folks like Cypress allow Matlab models to be compiled to silicon, to FPGA and SOC systems. If you need to provide a V&V record for your tools, as is true for any regulated, critical systems developer, you need Matlab-like, high quality tools.

So far as "Within the US, Canada, or EU" and "small bubble" point: true as far as it goes - but in terms of useful results, publications, and expertise you're not talking about a "small bubble" you're talking about the main course. In part because they start with these tools, and build on the work of several generations of research in their respective institutions. That conservation of effort, working with sound tools, allows a huge multiplication of effort. If you need to grow your own tools, it leaves you little time for interesting and innovative work in your field. Which by the way is a true principle in industry as well - it's almost always cheaper to buy your tools then to spend the time developing a "me too".

Lastly, you can talk about the "economic justice standpoint" - but the bottom line is that if you want to make progress, or make a profit, you use good tools which address your problems. Complaining doesn't make your tools better, nor in most cases, does writing a "one off" solution in the new hot language. Somebody, in this case MathWorks spend time and money creating and documenting a better tool. If I spend $10,000 to buy it, unless I can write it in less than a couple of months, I've saved money - that is economic justice. I suspect that with the variety of grants, discounts and gifts available there are few academics on the planet that can manage to provide similar tools in their departments.

A very silly story…

In 1980th I lived in a dormitory for post-doctoral students in Leningrad (Russia).

I had only programmable calculator. The winter was very cold and I was too lazy to go to the computer center. But I must to invert large matrix for Savitsky-Goley differentiation. I had no chance and I found the fast algorithm for generating the Savitsky-Goley digital filters by my calculator. If I had MATLAB…..

I believe it is our responsibility as programmers and researchers to come out with alternative open-source programming languages. Scilab is competing MATLAB nowadays, however, academic people still prefers MATLAB because it has powerful algorithms, friendly interface, and easy to use functions.

I hope that MATLAB would consider reducing its license more than it is nowadays especially for academic people.

In this connection:

Before you do the calculations you must think about "the use is going to be made of

the answer" (R. W. Hamming)

I think the open source (matlab and R replacing) programming language to watch is Julia: http://julialang.org/.

Taken from julia's intro:

Julia is a high-level, high-performance dynamic programming language for technical computing, with syntax that is familiar to users of other technical computing environments. It provides a sophisticated compiler, distributed parallel execution, numerical accuracy, and an extensive mathematical function library. The library, largely written in Julia itself, also integrates mature, best-of-breed C and Fortran libraries for linear algebra, random number generation, signal processing, and string processing. In addition, the Julia developer community is contributing a number of external packages through Julia’s built-in package manager at a rapid pace.

Example syntax (which is similar to Matlab):

function mandel(z)

c = z

maxiter = 80

for n = 1:maxiter

if abs(z) > 2

return n-1

end

z = z^2 + c

end

return maxiter

end

Another point for Matlab is Simulink.

Is there an equivalent to Simulink in Python or other language? I know of Modelica, which is free for the basic library but you need to buy the more advance one.

Also do they have advance variable-step solver like the ode23t (my favorite as an electrical engineer)

SciLab has a free, open-source version of Simulink, but I'll be the first to say that it's not as nice as Simulink is. It has a lot of trouble with nested superblocks and generally lacks some functionality over MATLAB.

On the subject of the variable-step solvers, it's worth noting that Scipy actually contains a wider range of them that are more performant, though I agree that it's a nice feature to have in any language.

That would be necessary if Matlab wasn't one of the most easily cracked software. I mean, you will find at least at least ten time more Matlab users than licences in most countries.

May be free MATLAB for non-commercial use will be more useful ?

Student versions are $99 instructor versions are free, making these comparable to any of the microsoft tools, and on the order of the cost of textbooks for most courses anywhere in the world.

When choosing the programming language/environment, ease of use should not be your only concern.

Is the speed of execution essential? If yes, forget all script languages and use a language with a good optimizing compiler (e.g., C, C++, Fortran).

Is it a relatively small problem, perhaps of a standard type (optimizing a few parameters, solving an ordinary differential equation)? Do you have to do the calculations for a short period of time? Use something that is easy to learn and to handle.

Is it a complicated problem? Do you expect to write a couple of 1000 lines of code? Use a language that you can adapt to your problem (e.g., C++). This is particularly important if your code is to be used and maintained by more than 1 person over a couple of years. In such a case, you will find that some easy-to-use programming languages will get terribly in your way.

Is a calculator a barrier to development of numeracy?

Is a keyboard a barrier to development of the ability to write?

Is a book a barrier to development of the ability of self-cognition of the nature?

The answer depends on how you use the tool.

Calculator, keyboard , book and MATLAB are only tools.

Blind use of Matlab harms as taking improper medicines.

I'd say that working with MATLAB over 10 years for scientific computation that MATLAB is not moving towards the right direction.  To me MATLAB is an inverted pyramid, whose foundation is pointy and the top is broad.  it's bound to topple.  Basically, MATLAB started as a Linear algebra calculator, then they started adding functionality, such as procedural and object-oriented programming, which comes out horrible for developing big scientific projects.  Bottom line, Python is the way to go in the future, there is a smart, capable, enthusiastic community behind Python development.  Very soon Python will sweep MATLAB's market.  And I will eagerly wait for that day.

Dear Srinivasan Rajaraman!

What we expect from MATLAB ?   User-friendliness and a wide range of applications. We do not care how it works, it is important whether it is useful to us.

Dear Slater Victoroff,

thank you for your question. I think MATLAB is only a mathematical unit assembly system. It is very large but not flexible. Therefore it is not suited for a new "architecture" of science. In technical science we need a simulator with linked parts of electrical network, state graph (machine) and information flow elements. Good systems simulators are Simplorer or Portunus. See my question: 

https://www.researchgate.net/post/What_is_the_best_development_language_for_new_computer_architecture_in_Micro-or_Nano-Chips

The idea of   "Good systems simulators" (Christian E. Jacob) is very interesting!

From computation to simulation!

Dear Joseph Dubrovkin:

Python now has user-friendly IDE.  One needs to get used to.  I use a scientific package from Python from Continuum analytics called Ananconda.  It comes with a native IDE called as Spyder, which is slowly but surely inching towards the MATLAB's IDE.  It's just getting used to my friend.  I am presuming that you may not have experience in developing algorithms in industry setting.  For academic, graduate research MATLAB is livable with, but dealing with high-volume data, MATLAB is abysmal.  

It just takes a little bit of resolve, effort, willingness, and urge to transition from MATLAB to Python.  I believe MATLAB is rendering upcoming computational scientists lame.  It is not stimulating for developing your programming skills in general.