/mlg/ Machine Learning General

>>61284114
>>
>a

>a

>>61284114
Past attempts at establishing a machine learning general have failed. /g/ just doesn't have the background and interest.

>>61284343
I'll just make this into README.md and publish on GitHub.

How strong of math do you need to really dig into this?

>>61284343
It's too bad, but /g/ has changed dramatically over the past 7 years. It's mostly just fags/kids with their consumer grade electronics.

>>61284366
Get a degree in Mathematics.
OCW is ok

>>61284343
>>61284374
I noticed this too, even though I hang around /diy/ more for actually technology related discussions. That part of 4chan actually talks about real technology and practical projects associated with them. Usually outsiders would expect /g/ place talking about design and development of electronic devices and applying science and tech in practical scenario but instead we get this endless scene of corporate worship of consumer tech and bait threads associated. I'm quite surprised there isn't a dedicated /ceg/ (Consumer Electronics General) sub board to help this board stay relevant to it's actual intent of discussing technology and design/development associated with it. I don't get it when /vg/ is created to mitigate the flooding of Videogames "insert game name here" General thread that /v/ once had, but /g/ doesn't get the same treatment despite all the consumer tech/goods generals thread this board gets daily.

>>61284358
https://github.com/Hypeouseaus/Machine-Learning-Guide-for-newbies

File: 1485724892782.jpg (11KB, 429x342px) Image search: [Google]

11KB, 429x342px

>>61284114
Okay, seriously how do I get started with Machine Learning...

I flunked out of the Coursera course after week 5 or 6 (when vision and neural networks started)

I get the hypothesis function and use a cost function to optimize variables to make the hypothesis function more accurate....

I also get that neural networks is basically simplified functions that output stuff that go into other functions...

literally can't make any other sense of it

Is there a for dummies book yet?

>>61285388
https://www.amazon.com/Machine-Learning-Dummies-John-Mueller/dp/1119245516

>>61285388
https://www.amazon.com/Machine-Learning-Absolute-Beginners-Introduction/dp/152095140X/ref=pd_lpo_sbs_14_t_0?_encoding=UTF8&psc=1&refRID=ZB5BEV2V6RRSY84RE0M2

Post speccy of your deep learning workstations.

>>61285451

Grasp how day-to-day activities are powered by machine learning
Learn to 'speak' certain languages, such as Python and R, to teach machines to perform pattern-oriented tasks and data analysis
Learn to code in R using R Studio
Find out how to code in Python using Anaconda

These are useless to me though, I know Python and care little for R ( I know C). I also already know wtf machine learning is...

>>61285388
I got one to thumb through while waiting for jenkins, it was pretty decent.

Called Machine Learning for dummies, just saw it at B&N while wandering on the weekend some time back.

Uses primarily python, worked at a reasonable pace, and did an ok job explaining. I have a problem of buying tech books and reading them, sometimes work will pay for them if I add them to the library when I'm done though.

File: yams.jpg (73KB, 640x640px) Image search: [Google]

73KB, 640x640px

>tfw in 1st year comp sci

i hope im not too late to ride the machine wave

Anyone else using ML methods to do quant things?

>>61285481
The suggestion stands, if you knew machine learning you would have passed the course but struggled. 99% of struggling in learning is because your foundation is missing something, perhaps a sentence slipped you or it was phrased in a way you couldn't make sense of but passed over it.

As as result when you get further on you are missing something and stuck. Coursera are pretty idiotproof and if you tried to dig around a bit you wouldn't be here asking for suggestions.

That said, the book is cheap, solid, and will bring you up to speed. Those points you mentioned from what I believe is the cover, are literally just advertising minor points, if that was the book it wouldn't be about machine learning, now would it?

File: Untitled.png (2KB, 233x74px) Image search: [Google]

2KB, 233x74px

How do you organize your deep learning papers? My backlog is ever growing.

>>61285501
Dang, way to make a person interested feel stupid - guess this shit ain't for me if there are people who don't struggle with the course.

I struggled at the coding portions - the quizzes were easy enough. But the assignments in Octave was loosely based on the concepts from the course and more to do with vector manipulations

The points I mentioned are the summary from the description... ie: the thing that is suppose to convince people to buy it. If you say it's more than that then sure... I'll give it a gander.

>>61285550
I didn't say anything about your intelligence, just persistence or dedication.

Also if you're struggling with the coding sections of that course then I suggest you get a bit more practice, and I say this not as a dick insult but as someone who wants to help you. I wouldn't respond if I didn't want to.

The course wasn't high level, it only expects as much as a college grad could do. Entry level pretty much.

Read the Ian Goodfellow book if you already have a solid undergraduate level CS and math background.

>>61285546
Can you share it? Torrent or 4lpha.one or google drive, anyways.

>>61285589
gotcha, can I just ask one thing then - is vector manipulations really that big of a thing for machine learning?

I stopped trying simply due to that alone. I never used linear algebra in industry after graduation (about 4 years since) so I'll probably have to retake a course on it - granted my dislike of matrix manipulation is probably strong enough to make me not learn how to do Machine Learning

>>61285546
Can you zip them up and put them on mega please.

>>61285625
I only save the ones relevant to my research/project so it could be biased. It's mostly arXiv papers posted on /r/ml so I don't think it would be of more use than things like http://www.arxiv-sanity.com/ where you can tailor to your need

>>61285628
You may want to brush up on your foundation, then try again. What you're describing are pretty basic in any type of programming outside of code monkey jobs.

What side projects are you guys doing in addition to your research? I'm annotating data to train an object detection model that identifies intimate interracial relationships in photos on social media.

>>61285685
>What you're describing are pretty basic in any type of programming outside of code monkey jobs

Where exactly does Linear Algebra and Calculus come into play in programming except for data science??

>>61285723
Computer graphics

>>61285728
Which is a meme field now as half of the papers at SIGGRAPH this year are deep learning applications.

>>61285763
Why?

>>61285723
I work in the financial sector making and maintaining small applications which act as a middleware between what others do in desktop interfaces and a larger processing server and database system.

Perhaps I've been doing this so long I have forgotten what is easy. Just keep on pushing and you'll get somewhere. If you want it bad enough you'll get it. If you do not that you will find some other interesting fancy or programmatic status symbol.

>>61284114
bump for interest

>>61284114
Why can't machine learning into chess?

Top chess programs these days (The top three change around, but for the most part, Stockfish, Houdini and Komodo shit on everything but each other) still use more classical AI algorithms that require lots of domain knowledge.
Will neural networks/SVM's/etc. ever be as good at evaluating chess positions?

I understand that alpha-beta and other pruning techniques are better than monte carlo for the search because the search space is somewhat manageable (compared to something like Go), but for some reason the evaluation part feels like there shouldn't be a reason not to be as good. I mean, people can easily get a database of high quality games online to train on, etc.

What's the best way to study without forgetting things you learned?

>>61284114

http://course.fast.ai/

>>61284114
/g/ is for consumerism and fizzbuzzing.

Go to /sci/ if you want to talk about ML and related.

>>61288044
First of, modern chess AI engines are good enough to win a game against any player in the world, for a practical standpoint chess is already solved.

Also the good thing about chess is that we have 100% of the information so we can check the game by brute force.

Of course what we want is to find a winning strategy, that is a strategy that always wins if playing whites. The state space of chess is insanely big. The problem is not the theory. We know that there exists a deep neural net model that converges to a winning strategy for chess, we just can't find it with current computers.

File: protest-hands-up[1].jpg (58KB, 840x442px) Image search: [Google]

58KB, 840x442px

>>61288825
>pic related is /g/ after that post