Thread 16705842

Serious question. Why is there no good statistics book that has enough math (probability and proofs), practical examples & problems (with data sets. programming), materials relevant to modern applications (machine learning, physics, etc.)? None of them is comprehensive, and pedagogically sound either imho.
I came from math and physics background, and statistics textbooks suck by comparison.

>>16705842 (OP)
because statistics is a weird field where there’s a giant chasm between theoretical statistics and applied statistics. You average “pure” statistician knows functional analysis like it’s his birthday and is so deep into autism that he will struggle to communicate it to the applied people. On the other hand, the applied people, due to how universal statistics is in other areas of life, are for the most part drooling retards who need to be told what a derivative is every time it’s brought up.

I prefer traditional unmathematical statistics.

>>16705842 (OP)
Because they don't know what they're doing either. Case in point, those free Springer Statistical Learning books. Literal nonsense. Or that Computer Age Statistical Inference. Hardly any logic in there. Just a bunch of techniques, and definitely not self-contained.

You need to learn theory before you can apply it. Machine learning is not too hard to grasp if you can understand linear regression and partial derivatives. The fundamentals of physics as well like motion can be understood with calculus. Calculating the rate at which something changes.

>>16707550
Learn how a really small neural network works and you can wrap your head around it. It is just inputs and outputs that are normalized. Even then, when you are building the kind of models used today all the math is contained in a black box understanding with machine learning libraries. Someone with no understanding of the higher level math can technically import some libraries and write a neural network with built-in functions and not many lines of code.

>>16705842 (OP)
Statistics are gay and probability is fake. All was decided by ALLAH(swt) at the start of excistence. Instead of participating in your cold cult of materialism humble yourself and submit to his will that was revealed by Muhammed(pbuh).

>>16707552
The main goal is to reduce error, loss. By using a partial derivative, we can get the slope of just one variable of thousands and figure out the direction and amount to adjust it to reduce the overall loss function. If we do this to every variable many times, which requires a lot of computing power, then we gradually get to a low enough level of loss where accuracy is reliable.

>>16707555

I am a Muslim and modern mathematics depends on Islamic inventions like algorithms and algebra. It is one of the reasons I love math.

You have a narrow understanding of both my religion and mathematics. Without Muslims, Newton's calculus would not be possible and neither would modern physics. This depends on algebra.

Without Muslims, we would not have algorithms and thus computers and things like machine learning and artificial intelligence. You project your own igorance on Muslims while nations like Iran have a very high scientific output.

Even Shariah governments understand the importance of investing in science and education, despite heavy sanctions, as art and science are the building blocks of human civilization.

>>16707567
>Without Muslims, Newton's calculus would not be possible and neither would modern physics. This depends on algebra.

>>16707567
Prove it by building a new Islamic golden age in the countries you are already a majority. I'll convert if, and only if, you do that.

Are you guys AI? The replies are so strange.

>>16707567
>while nations like Iran have a very high scientific output.
>'muslims' still being carried by their heretic-sect Persian scholars
some things never change

>>16705842 (OP)
There isn't a single book in any topic including Physics which is rigorous but also practical. The Venn diagram of people interested in either is disjoint.

>>16707550
Nope. Machine Learning really shines when you are talking about infinite dimensional spaces, for which you would need functional analysis.

>>16708885
>more layers but now with some arbitrary topology and a Banach norm
dear god…

>>16708094
If your conversion is motivated by the fact that xyz country is technologically advanced it probably doesnt count

>>16705853
>the applied people, due to how universal statistics is in other areas of life, are for the most part drooling retards who need to be told what a derivative is every time it’s brought up.

That is not even close to the truth.

>>16708897
Try teaching to a bunch of social “science” or business students, anon.

>>16708885
No one is looking for rigor. Just logical reasoning. Statistics books are often simply a collection of techniques that magically work. In a very disjointed presentation to boot. Not something you build up chapter by chapter. I’m sure there are many physics books like this too, but the good ones have enough math and logical sequence when introducing the topics, unlike statistics texts.

>>16708918
And surely asking for programming and data analysis exercises that use the online open data sets, is not much to ask?

>>16708920
Like check computational physics or scientific computing books. Or even statistics for physics books. They introduce their topics logically, with programming exercises, and have enough math.

>>16708885
What do you mean ML shines with infinite dimensional spaces? The majority of ML research requires, in terms of math background, not much more than standard linear algebra, basic probability/stats concepts and being able to do differential calculus with matrices.

I would agree that things can get more interesting when considering infinite dimensional spaces like function spaces or spaces of measures, along with functional analysis some more machinery such as convex analysis or optimal transport or calculus of variations might be used in these cases for example, but this is a small part of the ML literature.

>>16705842 (OP)
Tool builder or tool user. Pick one.

>>16706529
90% of people do.