Every machine learning project is written in Python

>every machine learning project is written in Python
>there are literally zero compiled languages with reasonable ML libraries
>most ML projects have abysmal code quality

Attached: 6e7.png (500x500, 201K)

Other urls found in this thread:

tensorflow.org/swift/
tensorflow.org/install/lang_go
twitter.com/SFWRedditImages

Why don’t you write one yourself?

ai wanna fug ai

I'm not capable of that user. I don't even write ML projects, but I would like to. But I write something that requires users to download a huge fucking runtime onto thier system which pollutes more than Chernobyl. Either that or you pack the huge fucking runtime into a binary but it'll end up being 100MB+.
Fucking clown world. Remember, these ML (((experts))) get paid 300K/yr to write code in a toy language with a disgusting runtime that'll be running in critical situations (cars, cameras, etc)

Because the important part is the training not performant code.

some ML is on the JVM

Pretty sure all the core tools are written in C and as such are not terribly slow.
Of course, a proper implementation would be to use C++ and offload everything to GPU, but still.

I said nothing about performance. The intensive stuff is done in a lower level language anyway and Python uses the FFI. What DOES matter is reliability. How the fuck can you rely on a language that can crash at any time because you passed a parameter of the wrong type to a function? This shit SHOULD NOT be running in cars.

Every ML library is C/C++.
Every ML Python project uses these libraries.

The reason is pretty straightforward.
You need the ML calculations to be fast - that's why something lower level like C/C++ that can be super optimized.
You don't, however, need super speed to USE these calculations. You want to calculate some shit and display it. Python is fine for that.

>This shit SHOULD NOT be running in cars.
I work for Ford. We don't run Python anywhere in the car - it's all C++.

OP that's because they're all backed by C++ libraries that do the heavy lifting. I get your aversion to python though.
>I'm concerned about the python frontend being bloated
It doesn't even compare to he libraries they use. It's crazy.

>I'm not capable of that user. I don't even write ML projects, but I would like to.

Yet...

>Remember, these ML (((experts))) get paid 300K/yr to write code in a toy language with a disgusting runtime that'll be running in critical situations (cars, cameras, etc)

Jow Forums in a nut shell.

Because ML libraries need to be easily tweeked.
The xkcd comic is not a lie. You mess around until it starts spitting out stuff that makes sense.

tensorflow.org/swift/

If every library is C++ at its base couldn't you just use C++ instead of python?

>reddit spacing
>completely missing the point
>Every ML library is C/C++
If that's true, where are the interfaces for other languages? That should be possible, yet it doesn't seem to exist.

>If that's true, where are the interfaces for other languages?
On their Github.
Tensorflow has official wrappers for Python, Java, and Go

shit, you're right. I had no idea that these bindings existed.
tensorflow.org/install/lang_go
thanks a lot.

>op didn't even search tensorflow language bindings
He won't make it.

it's in C++ so any language that supports interfacing with native libraries will work.
That's why they have Python, Java, and Go.

retard

VW is still a C shop.

>tensorflow.org/swift/
Still python.

>I'm going to do 0 research and complain on Jow Forums

I mean I don't know why I even expect some decency in this place.

Attached: 1543246149465.png (711x664, 314K)

Where I work every ML algorithm is done in OCaml, because you need certainty in your data (like when you have a dataset that declares itself valid even though it's clearly received after trading has closed). So don't just accept everything what you see in BEGINNER ML courses. People are doing all kinds of interesting things, since the 1980s, in other languages and sometimes even custom, probability or even process calculi type langs to do machine learning over massive data.

In fact this is how you best learn. Pick the language you like the most, do a 'data science' or 'machine learning' course and just translate everything into your own libraries. Some you can't easily do (visualization) but everything else you def can.

Tesla
C++ with a pinch of C#. Nowhere near the AP though.

Python is glue. Numpy, Pandas, Scipy, Tensorflow, Scikit-learn and so on are written on C, C++ and Fortran, Python only glue them together. At the end you get C speed minus a little difference that you will never notice unless you run a code that takes months to finish but at that point you (actually, your company) just run the project on whatever cloud service you like or at least, on a local cluster.

These are the type of people that bitch about programming languages, ladies and gentlemen.
They bitch all day about who uses what programming language but when someone asks them to actually do something about it, they let loose that they actually don't know how to do anything and base their opinions off of Jow Forums bait threads.
>why are machine learning projects written in python?
>well what's your alternative programming language
>i don't know...
>well are you working on an ml project in another language?
>I don't even write ML projects, but I would like to...

So is BMW/Mini/RR

Machine learning algorithmic developer here (usedto be a cosmologist). Python is the go-to tool here. My thesis was on the Milky Way galaxy and I do simulations of gas and how it moves around. I run simulations using C++ because its quite a fast programming language so it’s useful, but it's also much harder to write and less user-friendly than Python! So, I use Python for the more straightforward things, such as analysing simulations and ongoing analysis.

Attached: 1830.1549055226.jpg (800x451, 73K)

ml.net fag

Also, CoreML on macOS

>Muh meme learning
Pythonfags need to be lynched. Machine learning is a passing fad, no one will care about it and your retarded little SCRIPTING language (note that I refuse to call it a programming language) in a decade from now. C\C++ all the way.

>t. Embedded engineer

>t. "data" "scientist"

I'm so sick and tired that almost all research in NLP, CV and cognitive science is just training nerel netwerks to be 0.2% better than the old one without any insights on how on why and that so much human energy is used to train a systems that predict that if I bought shoes online then I should see ads for the same shoes the next 3 weeks.

The current ML wave really manages it to make the world a worse place in every regard it is used, I can't wait until they realize that they can't certify that shit and the next AI winter puts them out of work.

Also fuck python, at least use a real language

If it was only that.
> study machine learning
> start looking for jobs
> cam surveillance, detecting "toxic" behaviour, ad tech garbage

I like python though.

Sure dude, whatever makes you feel better about using crappy toy learning languages with abysmal performance. (((ref counting gc :sick:)))