Machine Learning

statistics and linear algebra are a must.
I would recommend you to graba good book on LA too, because you probably don't know as much as you think.

Use The Elements of Statistical Learning if you are a true beginner and have ok maths knowledgd

Are you a legit brainlet? Just learn R, C++ and Python, some Linear Algebra (you dont even need more than the basics), then go through the most used NN packages for each language. Make sure you understand everything you do, and if you need extra skills to implement what you're learning, learn them.

I know really only the básica on LA. Any particular book you would recommend?

And thank you all for the answers!

That's a doubt I have: I sometimes get the impression that there is not really much advantage in learning the theory behind Machine Learning, since the libraries and algorithims avaiable today are all relatively user friendly and easy to implement. Why should one pursue advanced learning in this field? Do the good Data Scientists out there actually use the theory somehow, or is it only knowing How to use the already given libraries?

PhD here

1: Learn linear algebra, multivariate calculus, multivariate stats/prob

2: Read The Elements Of Statistical Leabring 5 times until you get all of it

3: Take Andrew Ngs coursera course and be sure to do all examples in python, matlab, and r (yes all three of then)

3: Read this deeplearningbook.org/front_matter.pdf

4: Download tensor flow and do the examples

5: Go to arxiv and start reading papers and keeping up. Things change weekly

6: Start padding resume with jagger

Congratz you are now more qualified for a ML job than 99% of graduates, and are probobly more likely to get hired as well assuming you haven't killed yourself. Enjoy 6 figures and playing god

Fucking autocorrect

I meant pad resume with kaggle competitions

Thank you so much!

Just a couple of doubts:

1) What specifically about LA and Multi Calculus would you say is important to know? Any good books you can recommend?

2) The Elements book doesn't have a solution manual, which os really bad for me, who Will have to be studying on my own for most of next year. The Pattern Recognition one at least has some answers.

(What is your PHD on, by the way?)

Can you answer this pls Brainlet here. Why should I learn all that if using the avaiable libraries is so much easier, while yielding the same results.

Linear Algebra Done Right, Linear Algebra Done Wrong.
that Hirsch & Smale book is suppossedly good too, although more ODE oriented.

But anyway there are a myriad of books for LA, just google "best linear algebra book" and read stackexchange or quora answers, amazon users opinions and sometimes reddit is helpful too.
It's what I do when I wanna get a book on some topic.

What books did you use for linear algebra, multi variable calc, stats/prob?

1: You'll need multivariable calc for gradient decent. Linear algebra you'll use pretty much all of it. Just get a good book (I know it will trigger Veeky Forums but I recommend books for normies like Stewart for calc and Strang for LA. They are much better for what you're doing)

2: Elements is a book on theory. Yeah it doesn't have solutions but you really gotta grind and it's worth it

Computer science PhD

Because you don't want to just blindly use a black box and constrain yourself. You'll also be unemployable if they ask you questions on LA and you just respond "oh I just let tensorflow do that lol"

I listed the math books above, for stats and prob desu Ive never taught them just check the sticky, whatever is there should be good

do you want to be a machine operator or the guy who builds machines?

This didn't make It any better.

Do you want to be a cuck or the Bull?

Pretty sure the cuck is the one forced to use microshits leftovers because they can't make tools themselves

All you need multi calc for is the gradient decent?

Well there are other reasons, all that make Ng and Elements easier to go through, but saying :"all you need multicallc for is gradient decent" is like saying "all you need your legs for are walking"

>gradient decent

lul

hehe

How much money do you make?

Bump!

You need to understand the tool to understand what kind of problems it would be suited for and what the problem might be if results are different than expected.
If you don't understand this from the beginning you're never gonna be able to do your dreamjob of optimizing ad exposure on facebook!

I know that was a joke, but what cool things could I do with ML?

Hell I don't know I only ever use the more basic stuff although I agree that ESL is a great book.

But I think the real cool stuff like flying cars and programming a machine that is smarter than the other guy's machine is off limits for retards like you.

rude

Does ML have anything to do with Monte Carlo Makov Chain methods? I'll have to learn It next semester, and was looking for a good book .

Yes

Bump!

Unless you get a job as a researcher, you will be doing the most brainlet shit ever

And how does Machine Learning: A Probabilistic Approach compare to Elements of Statistical Learning? Is the target reader different?

What do you mean?

>3: Read this deeplearningbook.org/front_matter.pdf
All I see is the bibliography

you're never going to actually read and do all the exercises from a 600+ page book so it doesn't make any difference to you.

You're probably a casual faggot like me who goes on wikipedia and googles topics and scans through the page and then googles books and reads reviews and looks at chapter titles and day dreams imagining that he understands everything therein and does this for multiple different critically-acclaimed books and tries to find all the feed back to decide which book is the best without having worked through any, then gets bored and picks another fad to do the same thing again .
maybe next time it will be quantitative trading. or maybe you've already done that one ?

this will apply to at least 10 of the people who read this post.

Me desu

he's jealous of ML

>i'm a failure and can't get anything done so everyone else is too!
kys

I know nothing about Machine Learning, but that's a good book cover.

guy coming from FEM (aka real analysis applied to physics) here.
Why is ML so trivial in comparison?

any job he wants
300k starting

Machine learning is a meme.

Since this seems to be a data science thread I'll ask here.
I want to learn data mining, what's the best book for that?

he set a lower bound of 10 people, that isn't everyone

>finite elements
>dude lmao just approximate derivatives with a difference

That's finite difference brainlet. Finite element/finite volume methods provide cell averages in the solution domain by using the fundamental theorem of calculus/Stokes' theorem

t. Brainlet Abaqus user

I also recommend this: deeplearningbook.org/

it's more about neural nets (obviously) but it does it in a very statistically motivated way, so it's a nice one to couple with bishop's one.

>brainlets: haha ML is such a meme brainlet field
>brainlets after they skim through Bishop: pic related

delet

If all of you physics/math/engineering majors are so smart and ML is so easy, why don't you enter the field and grab yourself one of those $200,000 paychecks?

S-stop

Too busy with my $300,000 one already ;^)

The purple book is a nice reference to classical statistics, it makes no fucking sense to try to learn machine learning and have no clue of the foundations. (You don't need to read it back to back, but at least be sure you know about MLEs, confidence intervals, hypothesis testing).

Bishop is an introductory machine learning book, it goes though all the basic methods fairly well.

However if you are just starting out, check Ng's coursera machine learning course. After that you can go back to the books to get better fundamentals.

Bishop book is good, but its outdated. When you get the hang of ML and deep learning, read the 'deep learning book' by ML god Ian Goodfellow, which is up-to-date. After that, read recent papers to acquire knowledge of all the new secret sauces people are using. That's what i did

Don't bother. AI is going to be solved in a few years, you won't be able to catch up.