3 ms·
tl;dr: (1) Author does not understand the role of research papers (2) Claims mathematical notation is more complicated than code and (3) Thinks ML is easy becau
by amit_m 13y ago
tl;dr: (1) Author does not understand the role of research papers (2) Claims mathematical notation is more complicated than code and (3) Thinks ML is easy because you can code the wrong algorithm in 40 lines of code.
I will reply to each of these points:
1. Research papers are meant to be read by researchers who are interested in advancing the state of the art. They are usually pretty bad introductory texts.
In particular, mathematical details regarding whether or not the space is closed, complete, convex, etc. are usually both irrelevant and incomprehensible to a practitioner but are essential to the inner workings of the mathematical proofs.
Practitioners who want to apply the classic algorithms should seek a good book, a wikipedia article, blog post or survey paper. Just about anything OTHER than a research paper would be more helpful.
2. Mathematical notation is difficult if you cannot read it, just like any programming language. Try learning to parse it! It's not that hard, really.
In cases where there is an equivalent piece of code implementing some computation, the mathematical notation is usually much shorter.
3. k-means is very simple, but its the wrong approach to this type of problem. There's an entire field called "recommender systems" with algorithms that would do a much better job here. Some of them are pretty simple too!
- Aardwolf 13y agoI'm pretty good at logic, problem solving, etc..., but do find parsing mathematical notation quite hard. Is there actually a good way to learn it? What I have most difficulty with is: it's not always clear which symbols/letters are knowns, which are unknowns, and which are values you choose yourself. Not all symbols/letters are always introduced, you sometimes have to guess what they are. Sometimes axes of graphs are not labeled. Sometimes explanation or examples for border cases are missing. And sometimes when in slides or so, the parsing of the mathematical formulas takes too much time compared to the speed, or, the memory of what was on previous slides fades away so the formula on a later slide using something from a previous one can no longer be parsed. Also when you need to program the [whatever is explained mathematically in a paper], then you have to tell the computer exactly how it works, for every edge case, while in math notation people can and will be inexact. Maybe there should be a compiler for math notation that gives an error if it's incomplete. :)
- amit_m 13y agoYou probably want to look at a couple of good undergrad textbooks (calculus, linear algebra, probability). The good textbooks explain the notation and have an index for all the symbols. Unfortunately, in most cases, you have to know a little bit about the field in order to be able to parse the notation. The upside is that having some background is pretty much a necessity to not screwing up when you try to implement some algorithm.