Deep learning thread

hi

>>57415607
Proof that you're a human beeing.

>>57417761
He solved captcha.

>>57415607
That third image from top is fucking hilarious.

Who's the guy in 5th image?

>>57418713
Me.

>>57418783
Confident look, fashionable glasses, enticing background.

9/10 would probably invite you to dinner.

File: 1451935205109.png (19KB, 646x393px) Image search: [Google]

19KB, 646x393px

I made a neural net that outputs garbage because I'm too dumb to learn gradient descent or backprop.
I think I'll just train it via GA's.

>>57419405
>gradient descent OR backprop
yeah, you're retarded alright.

>>57419537
Forgive me, I've only looked into this fairly recently.

what's the best library to use for python?
link?

can someone recommend me some good books for learning?

>>57420460
>Doing machine learning in Python
Truly stupid.

>>57422275
Isn't Python the industry standard, and the only language supported by all frameworks?

>>57422315
It's also slow as fuck.
I have no idea why people want to do complex computations in this shit language.

>>57422325
what are you talking about?

All the scientific python libraries are basically interfaces for C++ code. They are as fast as they get anon.

>>57422315
It is. But it's slow at fizzbuzz which is the most complicated program /g/ can actually write so they hate it.

>>57424126
Lmao

>>57418713
Yann LeCuck

>>57419598
You're not the only one. Sit down and actually read it. A prerequisite is that you understand derivates at the very minimum.

>>57422325
Is it really that slow? I study scientific computing and I know that some of my professors do their work with python. I know that parts of the programm are also written in fortran, but I had the feeling, that the knew 'standard' for scientific computing is python...

>>57424650
the definition of slow in the scientific projects is determined by deadlines.

If there are no deadlines, then it doesn't matter if you have to wait 25% more time if the code in python is easier to programm than in said c/c++ or whatever

>>57424650
Python is pretty slow even when compared to java. Just search up a performance comparison on number-heavy computations. Additionally, and correct me if I'm wrong, but java can do some nifty shit with OOP that I don't know if the more script-oriented python can.

Noob here,
My main language is C and I'm currently learning C++, I know python's syntax but I don't remember threading, list comprehension and anything else than a couple libs.
Should I learn python if I want to get into deep learning/genetic algorithms or I'm good with C/C++? Most people seem to use python and I dunno if C++ would just make it more complicated than it is in Python.

>>57424650
It's absolutely OK. You are most of the time better off spending 100 hours writing program in python that will run for 15 minutes than writing for 150 hours in Fortran to run it for 10 minutes. Unless you are writing something that will be runned daily hundereds of times, of course.

Remeber, computer time is cheap, yours is valuable.

>>57422325
You just use cython instead and make that shit go sanic fast.

>>57420460
tensorflow

>>57420460
Theano if you're using windows. CNTK probably sucks with python

File: 1477445494806.png (88KB, 336x365px) Image search: [Google]

88KB, 336x365px

>>57422325
>language determines speed :^)
Go back to writing assembly, Ivan.

>>57424886
Which takes longer? Writing it in Python and waiting on the program to finish or writing it in C++ and waiting a short amount of time?

>>57425550
python has oop

>>57425550
>ITT people who don't know what they're talking about opine

>>57415607
I want to get into this field. I have a masters in computer information systems with a focus in data science. Hard to get job interviews.

Please whelps?

File: keras-logo-small.jpg (30KB, 300x300px) Image search: [Google]

30KB, 300x300px

>>57420460
If you want a lot of control over your architecture, use TensorFlow. Otherwise use a library like Keras or Lasagna. Actually, Keras uses TensorFlow as its backend, so it still gives you a lot of control if you bother digging through the documentation and code.

>>57427196
>Theano
Enjoy writing out weight update code for every project.

>>57422275
>yfw you don't know shit about neural nets

Don't worry, anon. You're not alone. Barely anyone ITT knows what they're on about.

>>57415607
OP I am interested in this care to share more.

>>57427303
kek it's like you people on purpose write inefficient programs and then blame it on the language.

>>57424180
Relevant article

Person fails FizzBuzz (write all numbers from 1 to 100, but write Fizz for multiples of 3 instead, Buzz for 5, FizzBuzz for 15), because it is "OMG MATH.". Claims it's the fault of the employers for giving her problems that don't have "use case[s]"

https://css-tricks.com/tales-of-a-non-unicorn-a-story-about-the-trouble-with-job-titles-and-descriptions/

>>57425553
>Noob here,
>My main language is C
God damn I hate /g/.

Python is only the machine learning industry standard because a lot of people in the industry are too stupid/busy to learn a "real" language.

I've seen my share of PhDs wiring absolute messes of code (python is not good for large projects). They say they will never learn C++ because "python is easier, you don't have to fight with the compiler and the syntax is nicer"

I've seen quite a few startups founded by 5-10 PhDs who all only know python. Seems like a recipe for failure, to me.

R master race reporting in

Who here /scala/ for ML?

>>57427829
What is the ultimate language then, my lord?

>>57427878
the ultimate language for him is the one that performs fizzbuzz the fastest.

>>57427878
R

followed by Julia.

Only n00bs use anything else.

>>57427918
R is shit and you know it. Julia actually looks pretty interesting, never heard of it.

Is Python really that popular in the deep learning/machine learning fad?

>>57428070
yep

there is masters course here called cognitive science. A coop between medical, philosophical and computer faculties. They are using orange, a python project.

>>57428040
R is awesome. Sure it is slow, but it practically codes itself. It is so easy someone with no math background or even programming knowledge could make a neural net in 5 minutes that is better than something a applied math/CS PhD could make in python in 30 hours.

Julia might be better than R.

>>57428070
I believe google use python for all their deep learning snizzle

>>57428070
it's what got me into python and I really like it

>>57427918
>>57428040

Julia is not a mature language so you will run into issues with libraries etc. I use scala (I'm this fag >>57427855) for my ML needs, and I'm very happy with it.

>>57428098
I think i might check out Julia. Still not convinced on the R thing, or can see how it's any better than python.

>>57428098
kek R is awesome?
nah and since you didn't prove it with anything, neither will I

>>57428136
It is way better than python for machine learning. It takes a moment to get used to but it is naturally suited for deep learning, machine learning, etc. it just does all the heavy lifting for you.

It is why data scientist and machine learning is increasingly being done in R. I realize people who wasted their time learning Python are upset.

>>57428174
Guess your right: http://machinelearningmastery.com/best-programming-language-for-machine-learning/
hmm who know, i might check it out, but ill probably stick to python for the syntax

>>57428136
Python with Pandas/Numpy/Sklearn is basically an R imitation that is better than base R but miles behind tidyverse ecosystem.

>>57428232
I am not trolling. For Machien learning, data science, data analysis, R is much better than Python. This is from someone who knows both. people who talk shit on R have never actually used it and just get their opinions from forums and web discussions from biased python users.

I am not saying R is better than python for everything, R is useless outside of analysis. machine learning, etc. But for deep learning there is no competition R is the king right now.

>>57427291
More abstractions more bloat the slower it is :)

We use both R and Python in production. R code is just as slow, but more stable, concise and readable, it also has great tools for visualization and presenting results. For deep learning its always Python because it supports all major frameworks (as of now R only has tensorflow and mxnet).

>>57422275
>look ma', I posted it again
ML is not for plebeian programmers, you first need to be a math or engineering elite.
Most of us don't have the time to write 1000 LOC to do trivial stuff, every time

>>57427878
There is no ultimate language, python included.

Python is missing a lot of useful features. If you have a large project it's not a good choice IMO.

>>57428174
>data scientist and machine learning is increasingly being done in R.
i don't know what you're smoking, but python is the one replacing R more and more in science in general.

File: lua[1].gif (13KB, 256x256px) Image search: [Google]

13KB, 256x256px

>all these people talking about Python

Lua + Torch for deep learning friends

>>57428520

This. All the actual big players in deep learning use Lua.

Python/Rust/Julia are for toying around like kiddies.

>>57428519
not according to actual data analysis done in Python and R. Python shrinks while R grows.

R is not for all science. So I am sure in fields not related to data analysis, machine learning, etc. Python is beating since R isnt used for these.

>>57428520
>>57428539
kek this thread is derailing as always

soon you will all talk about perl

>>57427797
Noob into deep learning/genetic algorithms,etc Im pretty fluent in C.

>>57428539
oh the language that starts lists at 1?

Caffe is the only good one for deep throating

>>57428557
learning*

File: 1285109261593.jpg (23KB, 499x375px) Image search: [Google]

23KB, 499x375px

>>57428556

I'm not sure why this is a problem with you. It doesn't change the effectiveness of the language just because it doesn't do something like your holy grail does.

>>57428550
>mature deep learning library used by all the big players
>derailing

>>57428570
sorry
Caffe is the only good one for learning deep throating

>>57428556

>Lua
>lists

It's okay to admit you don't know anything about Lua, anon.

>>57428579
>mature deep learning
>arguing which language is better cause phds suck at others
>mature
>you're actually providing to the mature deep learning
>bad b8
>i r8 8/8

>>57428582
I've used caffe before, and you are correct.
it's shit show trying to make it work with GPU, but that said it's very powerful framework

>>57428556
That's the only right way.

>>57425553
First go program something that will still have your data after you turn it off and on again, kiddo.

>>57415607
So the mathematicians were right?

File: 1392693398622.jpg (108KB, 964x768px) Image search: [Google]

108KB, 964x768px

>>57428600

Do you not understand the meaning of the word mature in the context of software development?

>>57428658
but then again /g/ isn't software development. It's only shitbags who still haven't created the /g/ distro.