Scientific computing using D

Hi everyone, I use mainly Python for scientific computing and I want to switch to D but I can't find good libraries as replacement for pandas, matplotlib, scipy, ... Are there plans to make D better in this area?

On Tuesday, 17 March 2020 at 00:48:24 UTC, ahmat wrote: > Hi everyone, > I use mainly Python for scientific computing and I want to switch to D but I can't find good libraries as replacement for pandas, matplotlib, scipy, ... > Are there plans to make D better in this area? https://github.com/libmir/mir-algorithm https://github.com/libmir/mir-random https://github.com/libmir/mir-blas https://github.com/libmir/mir-lapack https://github.com/libmir/mir-optim https://github.com/libmir/mir-intergral Mir is not the libraries for scientific research on a PC as well as the whole D isn't the best choice for that. Python is better for this kind of stuff. If you just like D, it isn't a good reason to use it. Use D if you don't have other choices because of technical reasons. D offers what other languages can't and sometimes it is the only good choice an engineer has. Ilya

March 17, 2020

Re: Scientific computing using D

Posted by jmh530
in reply to 9il

Permalink

jmh530

Posted in reply to 9il

Permalink

On Tuesday, 17 March 2020 at 06:02:25 UTC, 9il wrote:
> [snip]
>
> Mir is not the libraries for scientific research on a PC as well as the whole D isn't the best choice for that. Python is better for this kind of stuff. If you just like D, it isn't a good reason to use it. Use D if you don't have other choices because of technical reasons. D offers what other languages can't and sometimes it is the only good choice an engineer has.
>
> Ilya

I think it is useful for anyone to think about their use case and whether the programming language is the best tool to use.

I am often more productive in Python/Matlab/R for small projects that do not take long to write and where the code does not need to run for a long time. Knowing nothing about what the OP is doing, I would say start there.

However, if they are doing something that starts to run into the limitations of those languages, maybe they are doing something that uses a lot of memory or the code takes a long time to run, then they should start thinking about what the bottlenecks are. Sometimes with python, you might find the the bottleneck is actually some C code that numpy is calling. In which case, the improvement in performance may not be so large by using D instead. Other times, you can use something like cython to improve performance. However, then you start mixing C and Python in your code and then you might wonder why not do it in C first?

I considered using D because I was writing code that took several hours to run and sometimes ran into memory problems. I also didn't want to use C, and I couldn't stand the template system in C++. It has been good to learn D. I think I've gotten a better handle on many other programming languages because of it. However, I'm not sure I've saved as much time as I expected.

On Tuesday, 17 March 2020 at 10:28:00 UTC, jmh530 wrote: > On Tuesday, 17 March 2020 at 06:02:25 UTC, 9il wrote: >> [...] > > I think it is useful for anyone to think about their use case and whether the programming language is the best tool to use. > > [...] Python is easy to use but difficult to debug, Cython is a nightmare, C++ have many good libraries but it’s kind of complex. I am still a beginner in D, but if writing high performance code in D is easy i am definitely going to drop Python.

Have you considered julia? I like D and even tried to sneak it into as many uni assignments as possible but the package ecosystem isn't really there. You will spend a considerable amount of time on non-scientific-computing tasks such as writing interfaces for other c libraries. Op di 17 mrt. 2020 om 01:50 schreef ahmat via Digitalmars-d < digitalmars-d@puremagic.com>: > Hi everyone, > I use mainly Python for scientific computing and I want to switch > to D but I can't find good libraries as replacement for pandas, > matplotlib, scipy, ... > Are there plans to make D better in this area? >

On Tuesday, 17 March 2020 at 06:02:25 UTC, 9il wrote: > On Tuesday, 17 March 2020 at 00:48:24 UTC, ahmat wrote: >> [...] > > https://github.com/libmir/mir-algorithm > https://github.com/libmir/mir-random > https://github.com/libmir/mir-blas > https://github.com/libmir/mir-lapack > https://github.com/libmir/mir-optim > https://github.com/libmir/mir-intergral > > Mir is not the libraries for scientific research on a PC as well as the whole D isn't the best choice for that. Python is better for this kind of stuff. If you just like D, it isn't a good reason to use it. Use D if you don't have other choices because of technical reasons. D offers what other languages can't and sometimes it is the only good choice an engineer has. > > Ilya Thanks you, How is Libmir in term of performance ?

On Tuesday, 17 March 2020 at 11:50:26 UTC, Ahmat wrote: > On Tuesday, 17 March 2020 at 06:02:25 UTC, 9il wrote: >> On Tuesday, 17 March 2020 at 00:48:24 UTC, ahmat wrote: >>> [...] >> >> https://github.com/libmir/mir-algorithm >> https://github.com/libmir/mir-random >> https://github.com/libmir/mir-blas >> https://github.com/libmir/mir-lapack >> https://github.com/libmir/mir-optim >> https://github.com/libmir/mir-intergral >> >> Mir is not the libraries for scientific research on a PC as well as the whole D isn't the best choice for that. Python is better for this kind of stuff. If you just like D, it isn't a good reason to use it. Use D if you don't have other choices because of technical reasons. D offers what other languages can't and sometimes it is the only good choice an engineer has. >> >> Ilya > > Thanks you, How is Libmir in term of performance ? When compiled with LDC Mir has the same performance as C/C++ code compiled with Clang/Clang++. The linear algebra performance depends on BLAS/Lapack library that is linked with.

On Tuesday, 17 March 2020 at 12:10:24 UTC, 9il wrote: > When compiled with LDC Mir has the same performance as C/C++ code compiled with Clang/Clang++. The linear algebra performance depends on BLAS/Lapack library that is linked with. What he is trying to say is: The performance is about the best mankind has developed so far, sometimes + some small epsilon which he it about to minimize.

Forums