3 ms·

Here is my story. We used Matlab, but then the polling of the license server got more and more aggressive(or we were unlucky). Just when you want to work on the

Here is my story. We used Matlab, but then the polling of the license server got more and more aggressive(or we were unlucky). Just when you want to work on the 8h flight the thing popping up and saying 'need to connect to the internet' is not fun. Now we had a volume license(I think). And it needs server access every 10 times you run a script. Try making software which wants to use the Ethernetport with something else, it is just annoying. So you pay a lot of money and get annoyed... Octave is slower than Matlab (at least last time, no loop acceleration), and no GUI library. So we went straight to python.

enriquto 6y ago
> Octave is slower than Matlab (at least last time, no loop acceleration), and no GUI library. So we went straight to python. But don't you find python slower? I prefer octave to python, in part, because the linear solvers are much faster (and the startup time, too).
- andi999 6y ago
  To get fast you need to vectorize. I found that vectorizing algorithms in Matlab makes them unmaintainable (maybe I am not good at that). In python this is much smoother.
  - enriquto 6y ago
    > I found that vectorizing algorithms in Matlab makes them unmaintainable. In python this is much smoother. Are you using numpy? If so, can you give a concrete example?, in my case the opposite seems to be true. The numpy notation seems to be much clunkier. Apart from that, this sentence: > To get fast you need to vectorize. describes a very sad tragedy of numerical computing. It makes no sense that in 2021 interpreters for scientific languages are not able to optimize loops. The so-called "vectorization" should be a non-issue in a modern interpreted language. The running time of calling "C=A*B" and that of the triple loop to fill the elements of the matrix C should be identical.
    - andi999 6y ago
      I hope I remember these things correct. Let's assume you have two 3D arrays and you want to form the point wise product and sum over the middle index, you can probably write sum(a*b, 1). Anything more complicated there is numpy einsum (check it out) https://numpy.org/doc/stable/reference/generated/numpy.einsum.html https://numpy.org/doc/stable/reference/generated/numpy.einsu... Letme know how these things can be nicely done in Matlab. One problem I had in Matlab is some vectorization requires to permute the indices, but this is expensive if your array is hundreds of megabyte.

enriquto 6y ago

> Octave is slower than Matlab (at least last time, no loop acceleration), and no GUI library. So we went straight to python. But don't you find python slower? I prefer octave to python, in part, because the linear solvers are much faster (and the startup time, too).

andi999 6y ago
To get fast you need to vectorize. I found that vectorizing algorithms in Matlab makes them unmaintainable (maybe I am not good at that). In python this is much smoother.
- enriquto 6y ago
  > I found that vectorizing algorithms in Matlab makes them unmaintainable. In python this is much smoother. Are you using numpy? If so, can you give a concrete example?, in my case the opposite seems to be true. The numpy notation seems to be much clunkier. Apart from that, this sentence: > To get fast you need to vectorize. describes a very sad tragedy of numerical computing. It makes no sense that in 2021 interpreters for scientific languages are not able to optimize loops. The so-called "vectorization" should be a non-issue in a modern interpreted language. The running time of calling "C=A*B" and that of the triple loop to fill the elements of the matrix C should be identical.
  - andi999 6y ago
    I hope I remember these things correct. Let's assume you have two 3D arrays and you want to form the point wise product and sum over the middle index, you can probably write sum(a*b, 1). Anything more complicated there is numpy einsum (check it out) https://numpy.org/doc/stable/reference/generated/numpy.einsum.html https://numpy.org/doc/stable/reference/generated/numpy.einsu... Letme know how these things can be nicely done in Matlab. One problem I had in Matlab is some vectorization requires to permute the indices, but this is expensive if your array is hundreds of megabyte.