PyTorch: Where we are headed and why it looks a lot like Julia (but not exactly)

Post by laichzeit0 » Sat, Nov 27, 2021, 5:47 AM UTC

hjtkfkfmr wrote: post #66

forrestthewoods wrote: post #31

Earlier quoted context omitted.

I would claim the tech revolution happened despite those terrible languages rather than because of them. The languages are popular because of inertia, not because they're good. Python is popular because of the ML revolution. If ML didn't take off neither would Python's popularity. Is ML successful because of Python or despite Python? Well, the world is probably further along with Python than if it merely didn't exist…

As someone who uses Python since 2004, you got it backwards. Python was popular before because it's very nice language. People wanted to use it for science to, so they wrote very good scientific libraries for it. R was very popular for non-neural-network ML some years ago, yet it wasn't picked up for NN, because R kind of sucks for general programming. As the joke goes, the best part of R is that is a language writte…

Post by jhgb » Sat, Nov 27, 2021, 5:50 AM UTC

freemint wrote: post #13

adenozine wrote: post #5

Neither language have proper tail call elimination, which, is absolutely insane to me. Yall really just write procedural code for everything?

I've never seen recursive looping over a thing get turned into efficient SIMD code in any language. Starting with loops has no good reason to be better able to achieve that but for practical compilers it makes a huge difference. Julia code might also uses a lot of in place operations which would be hard for a compiler to infer as safe.

Post by chaxor » Sat, Nov 27, 2021, 6:04 AM UTC

Tarq0n wrote: post #130

xbpx wrote: post #46

Earlier quoted context omitted.

Another point that is often lost is how mathematical Julia code can look, especially for matrix functions. Yeh this is superficial, but so are 200 dollar sneakers and they do just fine. Honestly there is a real pleasure writing code that looks like it could be on the blackboard. The numpy / numba world in Python just feels... not great.

I actually consider that a negative. Mathematical notation is fine when you're dealing with limited space on a blackboard, but if you're programming please name your name variables and take the time to model your domain explicitly.

Post by dg96 » Sat, Nov 27, 2021, 6:10 AM UTC

nextos wrote: post #128

driscoll42 wrote: post #100

Earlier quoted context omitted.

Or I read it as "We want to make life as easy for our userbase as possible, so we will put more work on ourselves to make our users lives easier" which is an attitude I very much appreciate.

In the long run I think moving to Julia would make a lot of sense. I have used MATLAB, R, Python and Julia extensively for doing all sorts of data related things during the last 20 years. Julia is incredibly easy to work with, very elegant and really efficient. R and Python have always felt clumsy in some ways, and hard to write really performant code, even if you are more proficient in Python! As a seasoned Lisper a…

Post by jjoonathan » Sat, Nov 27, 2021, 6:25 AM UTC

rfw300 wrote: post #147

adgjlsfhk1 wrote: post #136

Earlier quoted context omitted.

Julia is generally in the same ballpark as Numba (depending on the application they should be within 2%). The difference is that Julia is a full language, while Numba breaks if you try to use it with anything else.

Yeah, Numba usually delivers on its performance promises when used right, but can be such a huge hassle oftentimes as to make it not really worth it.