Cython makes Python INSANELY FAST
Vložit
- čas přidán 22. 06. 2022
- Python has a bit of a reputation for being slow, but with just a little bit of effort, it can approach the sort of speeds you could expect from C. Wowsers.
-
If you enjoy my content, consider supporting me on Patreon (or hit the JOIN button below to become a member)!
patreon.carberra.xyz
If you need help with anything, feel free to join the Discord server:
discord.carberra.xyz
I get a lot of people asking, so here's my Visual Studio Code setup!
• My Visual Studio Code ...
-
If you have any questions, don't hesitate to ask in the comments! I'll try and answer as soon as I can, providing someone else hasn't already done so.
#python #coding #howto
Thanks to Mitko for pointing this out - due to an oversight in my implementation, the benchmarks for the factorial function are actually incorrect. This doesn't degrade the educational value with regard to the syntax, which is all fine, but I thought I'd just mention it. Moral of the story: be careful around big numbers!
Wait! You don't check if the answers are the same! I suspect if the code is converted to C/C++ without using Python's int object, it overflows and that's why it's way faster. It should be way faster eitherway, but does it properly use the python int object or just uses the primitive 4 bytes integer in C/C++?
You are...right. Fuck, how did I not notice that?
Thankfully I don't think it kills the educational value with regard to the syntax, but I will update the thumbnail to reflect the speedup from the sorting, which I've confirmed as legit.
Yo this is awesome bro, I run this code and it was actually 1867.571 times faster using Python 3.10.7 and C++ 14.0 (for the factorial() example)
Excellent video! New to Cython, but when I tried this without importing the cpython array it worked with similar speedups. Maybe they changed something in the last 6 months?
Entirely possible. I haven't really gone back to this since the video so I couldn't tell you. It's possible the speed was roughly the same anyways.
i'd assume the .so file is the same as every other .so file: a shared object. basically, a dynamically linked library.
Great video!
More videos on Cython please!
In the video you mentioned the pure python mode didn't give the level of speedup of full Cython syntax...I experienced the same. Is this behaviour to be expected? I would rather write full Cython syntax if that's the case.
Thank you! I believe so, as using Pure Python Mode requires the use of cpdef where cdef could be used in pure Cython. That's my understanding of it anyways, if I'm honest I haven't really done much with Cython since making this video!
You might write some Common Lisp params to benchmark them against Python programs solving the same problem. Common Lisp types dynamically. But some programmers wrote operating systems in it to run them on computers designed to run Lisp programs. Common Lisp lets you speed up your Lisp by statically typing some variables.
Can you call numpy functions from within a cython kernel?
"so" in "*.so" files stands for Shared Object.
I see -- I had a hunch but wasn't 100% sure. Thanks for confirming (:
so, is a shared object, something like *.dll in Windows platforms...
Can you call existing c math functions like matrix multiplication?
Yep. The documentation shows how, but I've never really tried it.
I wonder if PyPy could use some of its thorough data type analysis of Python code to convert Python to Cython. PyPy does JIT compilation but I wonder if mixing PyPy with Cython tools could lead to better performance.
how are classes with user functions represented in C?
no classes in c. use c++
Thx you.
You're welcome (:
You can't get 1000! fit in int64. Python's `int` contains all the digits, while C's `int` overflows silently.
I thought all the cython stuff in the .so file makes it so the int allocate more space
100,000! doesn’t blow up the max int value for C? That seems suspicious. Might want to verify that the outputs are the same.
naise
So confused. In the second example. Why did you have to make a separate cpdef list function ? I don’t see how it is even referenced from the first routine? I thought the whole idea was just to Cython existing code. This is very odd as it looks like you have to add a bunch of anscillary code to make it work and how would you even know what to put? My brain hurts now.
Use deques (double-ended queues) instaed of lists in python if one wants fast execution times. It's why they are a part of python.
I recently made a video talking about this, if you're interested.
Use numba in oython. Vastly simpler, and IT DOES GET ONE TO THE SPEEDS OF C.
Also, there's pypy3 and codon, a language tgat uses python syntax, but is compiled, so wicked fast. 😮
I've made videos on both Numba and PyPy, if you're interested.
Step 1: ask ChatGPT generate simple python program for hard task
Step 2: Cython generate C code
Step 3: Ask ChatGPT to simplify each Cython code files
Step 4: Go back to step 1
ChatGPT is not the end all be all. It's cool sure, it's not infallible.
Could you do a video on mypyc?
I certainly can! Planning on doing a video on that for next week (:
@@Carberra sweet!
Factorial 1000 will overflow your int. C doesn't care about the overflow, python does.
how can cython modules be pip installed?
Same as any other module.
@@Carberra And pip automatically builds the .so?
Cython builds the .so files. You then include the .so files in your distributed package, and pip will treat it like any other file.
@@Carberra and this makes your code no longer portable
Javascript also has to check types. Why is it faster?
Could you please make a series of Cython? It's really amazing. 💯
Thanksgiving.
If only I knew enough about it to say I was confident making a series on it 😅 I do keep meaning to go back to it and learn it more, just finding the time to do so atm is tough.
@@Carberra I can understand your hard work. Please proceed with the action in the future if feasible. Thank you.
😮😮😮
Oh, no. The narrator said, "can be RAN."
NO. CAN BE RUN. The word be is right there, so the past participle form must be used of the verb to run, not the past tense form of run.
Run, ran, run. It's not run, ran, ran. That woukd be incorrect.
Hell yeah!! Grammar Nazi, adore your
Spanish pliss
Its wuick, because it's compiled into machine code. Duh.
Bruh just paste the python code into gpt and ask it to turn it into pure c. Problem solved m
Learn to code. You can't rely on ChatGPT too much. Gemini is even worse.