3 ms·

Python vs. NumPy vs. Nim

18 points by narimiran 8y ago

narimiran 8y ago
Author here. If you have any questions and/or comments about Nim, I'm ready to answer.
- cicero 8y ago
  At this stage in my career, Python is my most used language, but I would prefer static typing (ideally with type inference) and native executables. I've been learning Haskell for that reason, but I see that Nim would also fulfill those requirements. How does Nim compare with Haskell, including its maturity and community?
  - narimiran 8y ago
    > How does Nim compare with Haskell, including its maturity and community? Nim is not as mature nor battle-tested as Haskell. Haskell has more users and it is more widely spread, so it would be a safe(r) bet. I'm also coming from Python, and I have tried several languages (including Haskell) until I kinda settled with Nim. I really like the similarities in the syntax between Nim and Python, which allows me to easily transfer my "Python thinking" into Nim code. There is also a possibility to only translate some slow/CPU-intensive parts of Python code to Nim, and then call it from Python (FFI). Regarding community - the Nim community (IRC/Gitter) has been very welcoming to me and to my beginner/dumb questions - I guess this is the advantage of a small community - there's a sense of a connection and people want to help each other.
amai 8y ago
What about Numba and Cython?
- narimiran 8y ago
  > What about Numba? I'll copy-paste my Reddit comment: Like pypy, numba was also on my radar, but I couldn't make it work by just slapping `@jit` decorator to the original code, and I didn't want to go into figuring out what's wrong nor making modifications to the original code. >> I suspect it should be on par with Nim. Somebody else [0] has run a comparison including Numba, where Numba is 1.5x faster than NumPy. This means, if we can compare tests on the different machines, that Nim should be about 2x faster than Numba. [0] https://github.com/henryiii/framework_compare/tree/master/lin_regression https://github.com/henryiii/framework_compare/tree/master/li... ---- > What about Cython? I haven't tried it. It would be interesting to see the results.
- henryiii 8y ago
  I've added several others, including C++, as well as a faster numpy implementation (avoiding mixing double and float): https://github.com/henryiii/framework_compare/tree/master/lin_regression https://github.com/henryiii/framework_compare/tree/master/li...

18 points by narimiran 8y ago

narimiran 8y ago

Author here. If you have any questions and/or comments about Nim, I'm ready to answer.

cicero 8y ago
At this stage in my career, Python is my most used language, but I would prefer static typing (ideally with type inference) and native executables. I've been learning Haskell for that reason, but I see that Nim would also fulfill those requirements. How does Nim compare with Haskell, including its maturity and community?
- narimiran 8y ago
  > How does Nim compare with Haskell, including its maturity and community? Nim is not as mature nor battle-tested as Haskell. Haskell has more users and it is more widely spread, so it would be a safe(r) bet. I'm also coming from Python, and I have tried several languages (including Haskell) until I kinda settled with Nim. I really like the similarities in the syntax between Nim and Python, which allows me to easily transfer my "Python thinking" into Nim code. There is also a possibility to only translate some slow/CPU-intensive parts of Python code to Nim, and then call it from Python (FFI). Regarding community - the Nim community (IRC/Gitter) has been very welcoming to me and to my beginner/dumb questions - I guess this is the advantage of a small community - there's a sense of a connection and people want to help each other.

amai 8y ago

What about Numba and Cython?

narimiran 8y ago
> What about Numba? I'll copy-paste my Reddit comment: Like pypy, numba was also on my radar, but I couldn't make it work by just slapping `@jit` decorator to the original code, and I didn't want to go into figuring out what's wrong nor making modifications to the original code. >> I suspect it should be on par with Nim. Somebody else [0] has run a comparison including Numba, where Numba is 1.5x faster than NumPy. This means, if we can compare tests on the different machines, that Nim should be about 2x faster than Numba. [0] https://github.com/henryiii/framework_compare/tree/master/lin_regression https://github.com/henryiii/framework_compare/tree/master/li... ---- > What about Cython? I haven't tried it. It would be interesting to see the results.
henryiii 8y ago
I've added several others, including C++, as well as a faster numpy implementation (avoiding mixing double and float): https://github.com/henryiii/framework_compare/tree/master/lin_regression https://github.com/henryiii/framework_compare/tree/master/li...