How did this SHIT language become the de facto standard for ML shit? Frick ython, frick pip, frick conda and everything that comes with it.
How did this SHIT language become the de facto standard for ML shit?
How did this SHIT language become the de facto standard for ML shit? Frick ython, frick pip, frick conda and everything that comes with it.
>filtered by python
LMAO!
The language itself is not the problem you moron. It's the broken tooling.
your broken tool aka brain is the problem
I don't think there's a single language with a good package manager anon
Cargo
u-uhmmm, Rustaceans, should we tell him?
NodeJS and NPM is infinitely better
>packages are per project project by default
>an actual package manager rather than a hack with environment variables
>node version updates rarely break anything (backwards compatibility)
>Node-API is API stable
>good tooling for native modules
>V8 is much faster than Python
post handwritten by a boomer, potentially an llm
Even NuGet is amazing compared to what we had in the '90s. (Nothing.)
node is great
simplicity is the best thing, you Black person
C
Maven just werks
sirs
Both npm (with or without yarn/pnpm) and cargo are better.
Pytards are wrong as usual
everytime i use npm i end up building a webapp, so that seems unlikely
The problem with Python is how hilariously slow it is compared to doing the same thing in just about any compiler based language. Do the same thing in C and it'll be 100x faster.
Do the same thing in lisp and it'd also be 100x faster. So what?
>muh c
go be cnile elsewhere please.
You see, software is a business. Its cheaper to just pay a bit more for the servers that run the software than to pay for the extra man hours required to develop something in C. This is even more true in software consulting.
>Its cheaper to just pay a bit more for the servers that run the software than to pay for the extra man hours required to develop something in C
shit like this is exactly why the free market can never work.
>shit like this is exactly why the free market can never work
leave it to the pants on head moronic commie homosexual to always post the most imbecile thing in a given thread.
>"shouldn't we refactor this so it won't become a problem down the line?"
>nah, that'd be expensive, we can always throw more resources at it (besides I won't be working here anymore when that happens)
that's the free market at work: people working according to their own self-interests (even if it results in a net loss)
Exactly, and there is nothing wrong with what you said. That's why I replied to the commie homosexual nonsense.
>will become a problem down the line
This is just your opinion and isn’t worth spending money on.
Deal with it.
real free market has never been tried
>shit like this is exactly why the free market can never work.
>that's the free market at work: people working according to their own self-interests (even if it results in a net loss)
But there was no reason to keep python so slow, especially when it's so fricking popular that guido is a billionaire. Developing a simple jit takes an afternoon and would speed it up 40x+ immediately.
Not enough profit incentive in doing so.
Quite the contrary, when it's not the people writing the software who need to do so (that's HOW people who make languages that become get rich: people contact them and pay them to implement features in the language, this is "consultancy"), can reduce server costs by 40x permanently, and further reduce overhead costs because you don't need as many IT staffs when you don't need the server scale you had.
no one likes paying upfront for invisible savings that might materialize later. deal with it.
(You) don't represent the rest of the world, thankfully.
I'm just saying it like it is.
You aren't. Otherwise, torvalds wouldn't be worth 500m and guido wouldn't be a billionaire and stroustrup would still be a pauper.
>Developing a simple jit
Like PyPy?
No, like numba but for the entire language. Pypy is a massive hackjob and supports frickall in practice thus any non-trivial code ends up equal or slower with pypy than cpython.
More like GraalVM with python.
>software is a business. Its cheaper to just pay a bit more for the servers that run the software than to pay for the extra man hours required to develop something in C
... or in Rust or in any other compiled language.
Why are compile gays so dense in the head about why people choose efficiency of human labor over efficiency of computer labor?
so much this.
I'd rather pay for an overpowered gpu once than waiting on some homosexual coding "elegantly" for who knows how long.
So don't do performance critical stuff in it? Wow so hard.
moronic take, since in shit like ml, you barely do shit in python, you just call into the c abi
cargo is way better, I hate npm, but it's still better than pip and can do local installations without playing with your environment variables
>cargo is way better
>i hate npm
these two statements are directly in contradiction, do you like npm or not
>do local installations without playing with your environment variables
why is that a problem? that's what environment variables are for
>interpreted is slower than compiled
STOP THE PRESSES
it's actually getting faster for a few years now, make sure you aren't using a slow ass outdated version
3.12 at the end of the year should fix the slow multithreading too
I want some of what you're smoking. Python 3 is still slower than 2.7 to this day. It's only getting slower if you use the nuFeatures and no faster if you don't. Also nobody cares that it's 500x slower than C instead of 502x slower.
you are the one smoking, python 3 is faster than python 2 since the 3.8 version, which is 4 years old.
and we have been getting big perf boost every major versions since 3.10, which was 2 years ago.
of course it's slower than C or whatever language gets you off, it should be obvious, but the fact is that it's getting really better.
Cope, seethe, dial 8 pytroony.
>python 3 is faster than python 2 since the 3.8 version
>shows python 2 vs python 3.7
you're so smart anon
Stop believing microbenchmarks. Just because it says micro doesn't mean it's as "good" (says you) as the dicks you like sucking on.
Then optimize moron
because it's the best C++ glue language, learnable by normies in a quick amount of time
people write high performance C/C++ libraries and then control them with python
This generation's BASIC, that's why.
Also this.
this is so fricking dumb
have you heard of lua?? that is the best glue languagw
Use Lua if you need a tiny easily-compiled runtime with as few libraries as possible. Use Python if you need the large and useful standard library conveniently available.
If you can't figure out that each one has a different general use case, you are fricking moronic.
>have you heard of lua??
The question is has anybody else heard of it. No and that's why Python is the go to.
abstraction power of python is superior
Its used by scientists a lot. Scientists make all the cool ML shit.
>ML
Ok, I'm already experienced with Python, where do I start with ML?
I know shit about math and statistics btw.
I want to use it to find patterns in financial markets, assume that I have the data, is it the right tool?
>I know shit about math and statistics btw.
Start there, then.
After that, andrew ng coursera, then deeplearning book, then random papers of interest.
>I want to use it to find patterns in financial markets, assume that I have the data, is it the right tool?
Basically yes, but actually no. Have fun(?).
>Start there, then
>good place to start
I don't think that I have time to learn math, if you have quick math resources that are tailored to understand ML do post them though.
Pic related is the peak of my math knowledge.
If only all the major languages and distros agreed on the same package manager standard...
>I know shit about math and statistics btw.
good place to start
>I want to use it to find patterns in financial markets
yeah...
please go be a CS undergrad somewhere else
I have a PhD in deep learning. You are a nocoder. You haven't even gotten to hello world yet.
would someone do that, just tell lies on the internet?
I don't know why it would be unbelievable, or even surprising. There are quite a few people beside me who also have PhDs, and deep learning has been very popular among tech careers for quite a while now. Combined, isn't obvious there would be some deep learning PhDs here? Answer: yes.
I don't doubt that there are PhDs on here, I do doubt that poster is one.
I know the mindset, I used to be like that, you lose it once you get your soul crushed in grad school/employment
Well, what can I say. Whatever helps you sleep at night, I guess.
>Ok, I'm already experienced with Python, where do I start with ML?
It depends, are you interested in actually having a basic idea of how the algorithms work or do you just want to learn how to plug a dataset into a python library and it does the thing you want?
>I want to use it to find patterns in financial markets, assume that I have the data, is it the right tool?
Yes ML is how you would tackle this problem, I have no idea how feasible it'd be to get accurate enough to make money though, I assume if it was more people would be doing it
Scientists tend to be shit programmers.
>Filtered by Python
BOT everyone
>engineer
t. underageb& nodev
AI winter would be my guess. by the time interest in AI was back all the languages who were used for it were long dead.
ML people are notoriously bad at programming and real computer science
this, when important shit does come out then you get stuff like llama.cpp where old school and sady dying off slavs and other remaining uncucked whites will port it to a non shit language
>ML people are notoriously bad at programming and real computer science
nuh-uh! ML IS heckin computer science. Programming just gets in the way.
> ML people are notoriously bad at programming and real computer science
you could say the same about software cooders and math, don’t get too wienery
People wrote software for python and nobody ported it to better languages. The end.
But all the good packages are coded in C and use Python as a skinsuit
Yes. Notice I said 'better' language, not 'worse'.
Would you rather be using R?
Python's popular data science libraries are mostly FORTRAN.
No, they're mostly C and C++. LAPACK and BLAS being used in the backend of THOSE C/C++ libraries is not relevant since they're switchable implementations, and especially since in practice we use CUDA versions for everything we can.
>Would you rather be using R?
Yes. Tidyverse is better than anything the pyBlack folk came up with.
better than node
Yup, but it does not make Python better
Researchers and scientists find python easy to learn and work with because it's easy to set up, the syntax is comfy, and there's a lot of support for math and machine learning libraries already there. Most of said libraries run C or C++ on the back end anyway for the number crunching. Almost nobody can be bothered to reimplement everything in lower level languages because whatever paradigm they're using will be outdated in 2 months anyway, by guess what? Another python model. Hope this clears things up for you.
He didn't say anything about lower level, though. For example, NYU used to use lush, a lisp language with neural network support, and pytorch used to be lua and used to be called torch. Porting effort to python started a little while before tensorflow, mostly motivated by theano. Then pytorch gained huge traction when theano died upon tensorflow's release, because tensorflow was outright unusable. Already people were deeply burned out on keras and how it changes everything every 5 seconds or implements layers wrong/backward with no docs on the subject; people were using lasagne with theano to avoid that crap. Then when theano died, people rushed to pytorch since it as the only reasonable framework left. Not before trying out scroogle's new trash, though. But I digress, the point is that it could have been lua, technically could have been lisp even though anti-parenthesis shills wouldn't hear of it, could even have been matlab or R (thank god it wasn't).
lisp autist detected
Try reading the post first at least.
excuse my moronation, but isnt Lua mainly for scripting within c++ programs, so said program is the caller? Unlike python scripts which are directly and call into c++ libs?
Just as js can be used for standalone programs, not merely as a scripting language in a web browser, so, too, can lua be used for standalone scripts (especially with luajit). Lua was indeed designed with good C/C++ embedding in mind, but that's hardly relevant, except inasmuch as you could have a C++ program for a "runtime environment" binary and use lua to manipulate a deep learning system that way, though that has never been done before.
lua is easy to embed into c++ but I don't you can easily call c++ library from external lua program without writing bindings.
python has some tooling to generate the bindings, some better some worse
>python has some tooling to generate the bindings, some better some worse
Not any more than lua does. You're thinking of C FFI's, which are all shit and broken, and aren't binding generators. A binding generator is something like SWIG or chicken scheme's EasyFFI (while an FFI, it does automatic binding generation behind the scene at compile time through macros as a second layer, which allows calling C functions without writing any glue code at all).
>Already people were deeply burned out on keras and how it changes everything every 5 seconds or implements layers wrong/backward with no docs on the subject
Examples?
why has noone made a compiled version of python? would fix all the speed issues
chat he doesn't know
Go back
no it wouldn't, look at pypy
>just in time compilation
not what i meant
Compiling or not is not inherently the problem. But the answer is that python is completely ad hoc, no spec, and it changes drastically between even minor versions, even at the bytecode level. Current techniques include jit compilation by e.g. pypy, a C-flavored python language called cython, that has support for interacting with python code, inline JIT compilation via the likes of numba, and VM-based fast execution with, for instance, graalvm. Numba is the only one that works consistently, but it only supports a small subset of python code. It JIT's the bytecode directly which is why it works so much better than the other approaches.
Python is good though, setuptools/pip is a clusterfrick, never used conda.
Cython.
Conda is infinitely nicer than pip+virtualenv, but still a clusterfrick.
You can't make Python a decent speed without breaking backwards compatibility. It's too dynamic and the object layout is absolute dogshit. The incredible inefficient object layout isn't actually specified in the Python standard but they still can't change it without breaking a ton of important C extensions.
>why has noone made a compiled version of python?
they have?
Is that a question or a statement?
If it's a statement, why the question mark?
because I'm confused as to why someone would be stupid enough to say "they should do X" without checking to see if they've done X. when a thought randomly pops into your head do you gays just go "damn I must be literally the first person to have ever thought of this, better not check though"?
>because I'm confused
Then make your statement and express your confusion, don't turn your statement into a question.
Try googling your stupid fricking thoughts next time anon.
I'm not the one who made the original post you replied to, moron.
I'm just commenting on your reddit speech pattern.
I'll defer to your expert opinion on how redditors talk.
"mypyc" does that, it compiles to C.
It work quite well but doesn't support all the langage. If you're already using a type checker for your script you should give it a try
because python is fantastic language in so many ways. only unemployed highschoolers larping as senior devs disagree
t. highschool dropout
t. CompSci high school senior without work experience
because ML/data science works primarily with iterable data structures and python syntax is convenient for that
L-lol?
Because it's easy syntax to learn and allows for quick prototyping. I didn't do a PhD so I could waste half my day optimising code.
Why is the Python environment so shit? Why does everyone tell you to use virtual environments? Out of all the other languages I've learned, only Python has this dumb shit.
Why do you think I created this thread. This shit triggers me.
Let me guess, the other languages you have used are PHP and Javascript.
Acshully it's C, C++, Rust, OCaml, Common LIsp, Racket, Chicken Scheme, OpenDylan, Delphi, C#, Java, aarch64 asm, and x86_64 asm.
>took the bait
So not only is it clear you are just a college kid who learned to compile hello world in half a dozen languages, you haven't used any scripting languages. So you don't understand the problem domain.
That's a lot of projection and self-own I see. Why are you so intimidated by the languages I have used for real world programming? Or, perhaps you're jealous that you are stuck with javas'crap at work?
reddit blackout really hitting you hard
Ruby was too good and beautiful for this clown world
Ruby's tooling is worse than Python's.
The only difference is that you don't have college dropouts with tiktok-fried brains trying to do advanced ML shit with Ruby, giving up after 15 minutes and starting threads about to cry about it like OP.
>tooling
what
be specific
gems, rbenv, rvm, all that shit.
Ruby dependency hell is the same as python's. In fact it's worse in my experience.
yah sure
bundler practically invented usable package management for a programming language
python still doesn't have a coherent package manager, it's been THREE DECADES
delusional
bundle install fails all the time thanks to shitty dependencies.
the problem with all these languages is the devs who write libraries and the devs who use them trying to cope for their abject failure to maintain stable APIs.
Perl invented real packages with perlmod. Ruby is an upgrade to perl and brought modules, aka gems, with it.
GEMS???
Ruby tooling isn't great but it's nowhere near as bad as Python. Bundler is de fact standard and better than any of the 27 different solutions Python has. Beyond package management, Ruby has better, nearly lisp-tier interactive debugging/programming/live-coding too. Python has... Jupyter lmao.
rbenv and environment managers kinda suck, I admit that, but they suck in Python too.
There is only one de facto package manager for python and it's pip. Even when using conda for environment management, you use pip for package management anyway, not conda.
It's true, though, jupyter is literally a hack to get lisp-like image-based development, but it's such a broken piece of shit and does basically everything wrong, it's hilarious.
>There is only one de facto package manager for python and it's pip
too bad real world don't work that way, for money you'll have to deal with everything non-defacto
because ML is for homosexuals who can't write real code.
>How did this SHIT language become the de facto standard for ML shit?
By figuring out how to make you seethe
read your post a few times and think about whether it makes sense
Just use Julia
install lua
even worse
I'm using Torch 6, stay mad
>arrays start at 1
Leave it to the Hues to come up with such blasphemy.
Anyone used to normal languages, will get hit by off by one bugs using Lua
A lot (not all) of pseudocode (and math) is 1-indexed, so that problem will always exist anyway.
THREE DECADES and python is still unusable
>clojure is a finished product
>ruby is getting performance improvements and has practially expanded onto BEAM and LLVM
>C89
>C++ 11
>Java 11
everything that's worth a damn has been usable for at least a decade now
only python, php, and javascript have coherency problems
small learning curve attracts least common denominator people and makes unusable shit popular
>How did this SHIT language become the de facto standard for ML shit?
AI and ML people are not the best coders do you know? Most don't even actually know how to code in a professional level, they know the equations and what they want to get done. Python is good for this.
The syntax is not bad, GIL is the cancer, need to use numpy to have half decent arrays is the proof that the core is corrupted. The language HAD potential, see codon for example, but the main stream use and libs in other languages make python a Frankenstein not a language.
How is the adoption of pyproject.toml doing in the real world?
https://blogs.gentoo.org/mgorny/2021/11/07/the-future-of-python-build-systems-and-gentoo/
Python has gained popularity as a programming language due to its simplicity, readability, and versatility. It is easy to learn and use, making it ideal for beginners and experienced programmers alike. Additionally, there are many libraries available for machine learning tasks such as scikit-learn, TensorFlow, Keras, PyTorch, etc., which make it easier to develop and deploy machine learning models. Furthermore, Python's integration with other tools like Jupyter Notebook makes data analysis and visualization more accessible than ever before. Overall, these factors have contributed to Python becoming the go-to language for machine learning projects.
Thank you GPT-kun, very cool.
I don't understand why Python became so popular either.
I can't think of a single situation I would prefer Python if I had the choice.
>I overdosed on Alan Kay writings and fell for the "dynamic types and message-passing language" meme
If you're one of these people you're on Ruby, Smalltalk, Lisp, Erlang
>I don't give a shit about the language, I just want a thin layer of C/C++ and the most moron-proof bindings possible
This is Lua
>I'm a system admin and want to shit out scripts to do my job
Perl is more unixy and, although people shit on the one-liners, they are copy-pasteable into a terminal window, which is nice if your an admin
>I'm a statsgay
R + tidyverse is better than anything Python has
>I'm an engineer
I hate to say it, but Matlab has all those proprietary drivers for lab equipment. You can plug it into any random oscilloscope and it just werks. The actual langauge kinda sucks but eng Chads are getting shit done, not jerk over language syntax.
So I don't get it. Maybe because Python would be like the 2-4th best option for all of these so it kind of has the closes average.
>lisp
>lua
>perl
>matlab
Are you delusional
Matlab is an abomination, but somehow you managed to not name R, which is even worse. Lisp and lua are god-tier. Lua even satisfies baby ducks like you.
>I can't think of a single situation I would prefer Python if I had the choice.
I use it as a shellscript replacement(most of the times) and I don't like Perl.
>coming up with increasingly delusional mental gymnastics so you can justify falling for another BOT meme
tiresome
no shit
It Hit the point of no return. Everyone was told it was the easiest language for newbies and lots of people stuck around and developed modules for it. Its just too big now.
Why python is shit:
1. Performance. wait till Black person business group writes code thats slow and offloads the processing to you because their Black person code doesnt scale
2. Documentation sucks ass
2.1 Core docs suck
2.2 "typing" always buried in modules. e.g., cant find the type of a boto3 object; need to grep through code to find types
2.3 Often wrong. e.g., websockets
2.4 every Black person and their Black person mother make blogs about how to do X problem shitting up web results
3. Linting is shit
3.1 e.g., websockets, AWS, json
3.2 50 different types Dict[] vs dict. Any vs any
3.3 ugly shitty typing syntax Black person[Black person]
3.4 often have to import types and do lib.homosexual_type
4. Libraries are shit
4.1 JSON has no object mapper
5. Error messages are shit, you get better errors in C
6. Many packages are incompatable with static analysis
6.1 dataclass_json
7. No common serialization and deserialization. You'd imagine a glue language would have a good way of transforming all data types to and from serial forms. No each lib has their own way and its all shit all the time.
8. len
9. pythonic. homosexual
10. whitespace Black person
10.1 copy paste code, too bad Black person cant auto format whitespace is syntax
11. package managing nothings good. everything is duct tape
Black person Black person only use this Black person lang because the libraries and ubiquitous Black person
>1. Performance. wait till Black person business group writes code thats slow and offloads the processing to you because their Black person code doesnt scale
don't hire Black folk and don't work with Black folk, simple as
>2. Documentation sucks ass
>2.1 Core docs suck
what's wrong with the docs? they're great
>2.2 "typing" always buried in modules. e.g., cant find the type of a boto3 object; need to grep through code to find types
>2.3 Often wrong. e.g., websockets
the frick are you typing for? memory is being dynamically allocated by the python interpreter anyway, you do realize types were originally made to tell the compiler how much memory needed to be allocated right?
>2.4 every Black person and their Black person mother make blogs about how to do X problem shitting up web results
more code to copy and paste from, not a problem
>3. Linting is shit
>3.1 e.g., websockets, AWS, json
what the frick are you linting for? Either the code runs or it doesn't
>3.2 50 different types Dict[] vs dict. Any vs any
actually every mutable type in python is just a wrapper around dicts, and almost all the immutable types you'll ever see are just the 4 built-ins (numbers, strings, tuples, frozen sets)
>3.3 ugly shitty typing syntax Black person[Black person]
Stop capitalizing your variable names you Black person
>3.4 often have to import types and do lib.homosexual_type
from lib import homosexual_type wasn't that hard
>4. Libraries are shit
>4.1 JSON has no object mapper
objects are just fancy wrappers for dicts, which JSON is.
>5. Error messages are shit, you get better errors in C
wtf you get a full stack trace it's great
>6. Many packages are incompatable with static analysis
stop doing static analysis Black person. why would you work to statisfy your compiler's arbitrary wants. if you wrote the code you know if it works or how to debug it
1/2
This is the worst bait I've ever seen.
>Python docs are good
>Static analysis is bad
>Just write unreadable code
>Just have 50% of your code be imports
Just look at regex docs for python and then compare it to C++. You have to read a homosexual ass essay to get any info
>Just look at regex docs for python and then compare it to C++. You have to read a homosexual ass essay to get any info
That's because you read the tutorial instead of the docs. The docs for regex are as straightforward as any other docs
docs are good
yes
analysis is bad
yes
>>Just write unreadable code
whitespace is the best thing for readability, as is keeping lines short and using descriptive variable names written in snake case
>>Just have 50% of your code be imports
yes
Anon, it's clearly not the worst bait you've ever seen since you're responding.
>6.1 dataclass_json
don't know what this is, don't care
>7. No common serialization and deserialization. You'd imagine a glue language would have a good way of transforming all data types to and from serial forms. No each lib has their own way and its all shit all the time.
pickle
>8. len
is this an objection? "oh I hate functions who's use is predictable"
>9. pythonic. homosexual
>10. whitespace Black person
pythonic show someone who knows what their code does.
whitespace is the ultimate way to enforce readability, that's why python doesn't give a shit about linting
>10.1 copy paste code, too bad Black person cant auto format whitespace is syntax
good, gives you a chance to read the code as you fix the whitespace
>11. package managing nothings good. everything is duct tape
pip is great
picrel is you with all your tools for "safe programming"
>cant find the type of a boto3 object
skill issue
there is literally nothing a wrong with the whitespace
you're using whitespace anyway to arrange your code anyway
The other one
>moron Oriented Programming enjoyers
are now all using Go. Dumb simple language with only one way to do things.
I liked Python 2.7 but they've added lots of shit to the language, it's not really a simple language anymore.
Seething comp "science" cuck filtered by Chad physicist invention yet again!
?
Physicists are the ones using python, which was made by a "cs cuck".
I thought physicists used c++ because of the parallel programming allowing to run wave simulations faster on GPU’s.
Physicists let the giga-autist numerical physicists do that and wrap it up for us in a nice python package
cniles had to pick up the next moronic thing next to them
I FRICKING HATE THE GIL I FRICKING HATE THE GIL I FRICKING HATE THE GIL!!!!! REEEEEEE
>J-JUST RELEASE IT OR USE MULTIPROCESSING
OKAY NOW EVERY. FRICKING. FUNCTION. NEEDS. TO. BE. PICKLEABLE.
I HATE THE GIL
>Python is b-ACK!
BACK
If I want to run a certain version of Python software, that runs only on a certain version of python and needs fixed versions of dependencies, I download all that shit (software, python and dependencies) and save it away so if I ever need to re-install that exact shit again all the files are ready to go. Anything else is a world of pain. Expecting the internet to save your shit for you is low effort.
conda and pytorch-lightning are cancer
the rest is fine
Pytorch-lightning is such a profoundly pajeety abomination, how can anyone unironically use that garbage is completely beyond me, but then again some people use tensorflow and keras.
It's such a cult. I remember pytorch-lightning's devs actively shilling their shit on reddit
What? You don't like a moronic abstraction layer that does not add anything to your model except for deprecated features? And moronic ass implementations of models that you can find everywhere else??
Honestly the team I joined used it for prototyping and even then I absolutely don't get the point. I moved everything back to pytorch, at least all of our production inference code.
I didnt witness the shilling, what were they even selling it as? Like why would pytorch need it's version of keras? Keras is only useful because tf is or was messy all around. Pytorch has its own moronic feature chasing and deprecation hamster wheel (only for non core features though), but it's super manageable still.
>I didnt witness the shilling, what were they even selling it as?
something something "if more people use it, we can accelerate ml development because of standardization" and then constantly fight back the arguments against pytorch-lighting on reddit
>t. nocoders allergic to getting shit done
Pytorch-lightning literally requires MORE boilerplate while being LESS flexible for the same thing. It does the precise opposite of "getting things done".
>Pytorch-lightning literally requires MORE boilerplate
Bullshit.
model = Autoencoder()
trainer = L.Trainer(accelerator="gpu", devices=4)
trainer.fit(model, mnist_train, mnist_val)
Let's see your alternative.
>while being LESS flexible for the same thing
Bullshit.
https://lightning.ai/docs/pytorch/stable/starter/introduction.html#maximize-flexibility
Hello, projecting looney troon. No wonder you talk like a predditor.
Who pays you to make these negative-effort shill posts, pajeet? I wouldn't even give you 0.02 rupees for this crap.
>t. model tinkertroony
You have never deployed a single useful model in prod, much less worked on a model that didn't come straight from some premade lightning bolts trash implementations lmfao. I'm sure it gets you to a pretrained yolo hello world equivalent very fast so it makes you feel you actually did something lol.
Imagine getting filtered by vanilla pytorch jfl hahahaha
having another dependency that adds little to no benefit is anti-pattern
what other languages have list comprehensions?
- List comprehensions are shit. Cruft at best.
- Haskell
Chris Lattner (from Clang/LLVM) and a bunch of Google and Facebook engineers are working on a new language called Mojo specifically for AI. Syntactically it's quite similar to Python but it makes much better decisions in terms of performance.
>The Mojo standard library, compiler, and runtime are not available for local development yet, so we created a hosted development environment where you can try it out. We call it the Mojo Playground!
You'll own nothing and be happy
It's moronic garbage.
>frick pip
homie you type "pip install library" and library installs
I love python, reading/writing it, pip, conda, environments...see github..I can quickly start working on it myself if I want. ez.
Want to know how I know this board is full of larpers. Not one person brought up the python REPL yet, the actual reason why you want to write ML code in python instead of C++ or some other compiled language
If the package manager is so bad why has no one written a good one?
theres poetry
what
would you rather they chose R?
R is actually pretty based but lacks scaling for real ML or computer vision use.
WHY do you hate python?
Frick python really. Frick your moronic whitespace sensitive shit. Frick your global lock. Frick your GC.
Python uses refcount, not GC. The GC is only relevant when you have cycles (which you never have).
nothing wrong with whitespace, moron
bump I didn't write all that shit out to fall off the page now
hating python is the ultimate midwit opinion. here's a list of people who hate python:
>gays who just decide to hate whatever is popular in lieu of actually knowing things or forming their own opinions
>unemployed nocoders who just repeat things they saw on 4chins
>people incapable of learning more than one language, or incapable of discerning what language they should use to approach a problem and think they'll just learn one universal language that will somehow be perfect for every application
>people who somehow got filtered by something simple like pip or somehow fricked up their environment
>people who have never written anything that needs to be performant or resource-constrained who've decided to larp as embedded programmers even though their shitty little script would have been 1000 times quicker to write in Python (but I needed this noncritical task that runs twice a day to complete in 1ms instead of 37ms because I JUST DO OKAY?!)
>geniuses that don't realize that the parts of their favorite library that actually need to be performant are written in C anyway
get a job you stupid fricking hippies
option D: people that have python shoehorned somewhere in their daily work but are experienced enough to understand it has no place there when better options exist
>option D
I worked on a 450k sloc Python monolith, which included a custom templating framework and an interpreter for a DSL for generating business reports.
I have PTSD from all that shit code and the """Pythonistas""" who wrote that fricking thing and were PROUD OF IT.
People who specialize in Python are legitimately bad developers and the language is a blight on the world.
That said, it's fine to use Python for throw-away shit you don't care about.
>Pythonista
Normies were a mistake.
>muh normies
If you would have worked in the software industry you'd know that most software devs are indeed normalgays with gf's/wives, children, friends, hobbies that don't involve computers etc.. This type of sentiment is often common among students and NEETs who think being able to program is somehow mutually exclusive with having a normal social life.
Maybe if you're a webshitter or """pythonista""".
I worked in embedded, not as a dev, but as a technician and later as an integration engineer which are basically the same. Dev teams there are literally the most normal people you'd see ever. The last team I've been a part of wouldn't look out of place managing a department store.
Nice cope. My company has projects in multiple fields (including but not limited to backend systems for financial software, enterprise CRM and embedded software for various industries) and employs around ~150 developers. The majority of them are what you'd call a "normie" (absolute reddit newbie lingo). Better accept the fact software dev isn't some social outcast clubhouse before you enter your first job, that way you aren't in for such a massive shock.
They didn't used to be. They did have wives and shit but they were inherently misfits, and all the people who participated in shit you've heard of had PhDs or dropped out of one (a fact I only learned about recently, even in videogames it was like that back in the ps1 era, let alone in OSs in the 30's to well into the 80's). By misfits I mean that they were almost all very eccentric characters with quirks and social disabilities profound enough to be noted by their coworkers.
Or, you know, people who actually have to use python for work.
If you work for a shitty company with incompetent leadership and lazy programmers, your codebase if going to be garbage no matter what the language (they were dumb enough to hire you, after all).
t. NEET
I'm sorry to hear about your moron coworkers but
is 100% right and literally any language will be hated if you're forced into using it for work by morons.
Of course he's right about that part, but it's completely irrelevant. You can't find a single working professional who likes using python. Are you saying that the only places that hire python users are shitty? Then that's a self-defeating angle.
>You can't find a single working professional who likes using python.
If you're going to larp as someone who has ever held a job above "fry cook" or "stockboy" then at least try to be more subtle about it.
I've worked for a decade as an embedded RTOS avionics software engineer and yet I love Python for basically every part of our environment/testing/build process that benefits from using a scripting language.
Oh, it's the resident schizo again. Should've known. Have you figured out how to not segfault your hello world program yet? Or are you still too busy fapping to MISRA-C to get there?
>Oh, it's the resident schizo again.
The fact that you've invented some other anon in your mind to project your illness onto says all I need to know about you. Take your meds and get an actual job that isn't "BOT shitposter".
Take your meds, schizo.
>no U and buzzwords
Stay poor and unemployed.
>You can't find a single working professional who likes using python
I can. YOU can't, because you don't know any working professionals, the only programmers you know are nobody gamedevs you follow on twitch.
Well then I'm waiting. Where is this magical python liker? I worked at microsoft, google, and a few startups you've never heard of. I've done a bit of everything except webtardation. I'm still waiting to find anyone who likes python.
>they are actually proud of being slaves
point and laugh
>You can't find a single working professional who likes using python.
who the frick is stupid enough to believe this. I don't even think YOU are stupid enough to believe that
t. NEET
Because the level of introspection and reflection it provides allows for writing idiot-proof code. Take the pad function of the Huggingface's SequenceFeatureExtractor, for example:
processed_features: Union[
BatchFeature,
List[BatchFeature],
Dict[str, BatchFeature],
Dict[str, List[BatchFeature]],
List[Dict[str, BatchFeature]],
],
This one can take a BatchFeature, which is basically a subclassed dict, but also either a tuple or a list of it, but then again many other nested collections. The method then does all sorts of transformations including on the arguments themselves, converts the underlying data (1D floats basically) to numpy arrays and calls another method, _pad, that makes them all the same rank by padding them with zeros or whatever the padding strategy is desired.
I've been translating it to Kotlin and dealing with Any? all the time is a pain in the ass (not to mention Kotlin's rather limited reflection capabilities) and not doing it the standard Java way (method overloading) makes for a very tedious work.
But for non-Python wizards, typing pip install transformers and importing Wav2VecProcessor and just shoving it with data and the library doing all the pseudo low level stuff means that pretty much everyone with a gram of brain and enough computing powah can train produce a deep learning model by writing just a few lines of Python code.
So in essence, the libraries and packages around ML shit are written for non-coders in sight and the language's design allows for that.
N.B. this is not a ChatGPT post although it might read like one.
it's almost like not doing tedious work and things working is... le good?
Oh yeah, I agree wholeheartedly, I love Python. I also like how Huggingface's transformers are written. They are over engineered for sure, but it certainly leads to less code because it is a dynamically typed language after all, when compared to Kotlin for example. Being able to allow the end user a tolerance for what they give your functions and adapting to it before you do the actual gist of preprocessing is very cool. And this tiny (compared to the actual number crunching underneath) overhead is of no particular meaning to the end user. I was just trying to give my impressions why Python is the de facto ML language.
>i will type many words but not say anything meaningful
ok here is something meaningful - it does not matter what language you use as long as you get something done with it. i and a lot of my peer can get a lot done with python.
You're not wrong. All things considered, it presently manages to beat all alternatives.
At least they didn't choose PHP or Javascript.
Look on the bright side: at least it's not javascript.
Python has some of the worst tooling i've seen in a semi-modern language and every python software is like 50% tooling.
>pip env, conda etc
Which is honestly the main reason why I think Go will outdo it in the long run. The tooling is good and running go software involves you typing "go run" instead of futzing about with environments and package managers.
Businesses don't know or care how fast their code is. The internal software at most companies is a complete fricking mess that will never be fixed because everyone's hiring pajeets to write godawful code and plug Microsoft Solutions into everything. At this point the pajeets are just functional randomware and the only way to sort the whole mess out would be to start from scratch with competent engineers, but THAT ain't happening.
Python is really popular with third worlders like indians and slavs.
School.
>favorite programming language of third worlders
>favorite programming language of beginners
>favorite programming language of career shifters/bootcampers
Gee I wonder why
that's JavaScript
you forgot:
>favorite programming language of geniuses and cool people
Name one cool person or one genius who likes python in any capacity. I'll wait.
Me, i'm cool as shit.
You're thinking of Jabba and Jabbascript
>open linkedin and search computer vision jobs
>they require c++
Why the frick do you even need it? I've never used c++ in my entire ML engineer career
Computer vision means old-style sift and hog and Fourier analysis, correlations and ad-hoc algos, not deep learning. Some people really hard 50 years behind the time and refuse to evolve.
>import jax.numpy
>I've never used c++ in my entire ML engineer career
>career
Copying code from youtube videos isn't engineering
Which part of your job needs custom c++ implementation?
Model training and external libraries.
You are so moronic that I'm getting second-hand embarrassment from your post.
Says the third worlder who didn't use C++ for ML. How's bootcamp going juan?
cringe
>look i'm a jobless NEET but at least I wrote my calculator in C for da real hackerz not in heckin python for kids
Python is a fricking overglorified TI-84 calculator. It should be used by noobs to learn how to code, script simple stuff or compute problems and uses like numpy and Pandas are ok in a scientific field.
But making websites and other enterprise level software with Python is a extremly stupid idea and everyone who insists on it deserves every error which occurs in Python. TypeError, Python 2 and 3 issues, optional strong typing, pip, you name it.
Python is quite good. It has a lot of libraries that help you build interesting stuff, and comfy frameworks that make it easy to put it on the web. With JavaScript, for example, you can only build generic type of sites.
Which AI did you use to generate this amazingly moronic post?
prove it wrong, gay
The language itself is fine, it's the tooling that's is garbage because it as mean to be a scripting language for glue code but now it has been lifted to the dignity of "full programming language" that it was never meant to have, and the PSF people are stuck in the 90s and delusional and refuse to standardize proper tooling
Do you use PyPy or Cython?
pypy never works and cython isn't python-compatible, so cpython.
It's easy to learn for non-programmers and scientists from other domains could easily start to use it and contribute to machine learning research. Is this really not obvious?
It does attract homosexuals because the learning curve is low. I know a guy who writes python scripts in layers upon layers of abstractions like java. Frick that guy
in the time it takes you to push your repos first commit, 20 new data quality management frameworks were released on pip, with 20 different factory method signatures that output 20 slightly differently wrapped namedtuples and you somehow need to test them all
youd best get abstracting boy
I don't mind proper abstraction, but doing it for the sake of "future proof" or "not abstract enough" is insane
indeed, you have actual ML languages like standard ml and ocaml, why go for python to do ML..??
python is based. pip install whatever in your venv/container works fine for people with jobs (well paid, btw)
the thing I hate though is that it follows 90s OOP rules, so lots of the most popular modules are implemented as single-method classes with like 10 stacks of inheritances that you have to scroll past in every stack trace
t. never used python
We could all be using Nim
sex-havers use python.
When you open a console in a video game, do you also seethe that it's written in lua?
Python is the public API for a shitload of c/c++ libs. If you don't want to use it, then you're more than welcome to write your own interface in whatever lang you please.
>Protip: you won't because you can't