lindbergh

More details on the math related to it than you'd like : https://www.youtube.com/watch?v=OjcacSB6_EE

Right, good point. I just find the structure and the mindset to be similar.

In their definition of programming languages types, I'm not sure I see how SQL which they cateogrize as declarative really differs from FP, especially with lazy evaluation. Both are a succession of functions applied onto a previously declared variable : just think of a long SQL query with many ctes modifying the previous ones; each of them can be thought of as a variable which takes its value from the application of a function on other variable (a bit like the let function in ocaml).

I always thought of LaTeX as being the prime example of declarative programming.

Oh wow that's really impressive. Thanks a lot !

Yes Excel lately is turning into a really compelling product, using functions such as sequence (dynamic arrays), lambda (pure functions) and let (defining sub-variables). You'd think you're coding in a pure functional language, and that's great! Now, ff only Excel had a convolution function! But nonetheless it's a great GUI, everybody knows how to use it and sharing it is dead simple.

The one downside I can find is the lack of a good plotting library. And yes comments as well is something I miss a lot.

Interesting. I heard many good things about Pycharm, but it kind of looks bloated? How does it compare to VS Code (which is quite good (notebook and repl support mostly), but not great, in particular regarding the debugging experience)?

What I wish more newspaper did is to allow one to pay for a single article (whose price could depend on the length/depth of the article).

Recently stumbled upon this quite good YT channel, which also brands itself as aimed towards beginners, although in my opinion both the Susskind lectures and these 'physics mini lessons' would be best appreciated with at least some prior knowledge of undergrad physics.

I really miss doing physics!

https://www.youtube.com/c/PhysicswithElliot

Well you can manage to run the Wolfram Engine through a jupyter kernel directly in VS Code. Although, I believe when I set it up the output was an image, no matter the object being returned (for instance 2+3 would return an image of 5 black on white).

Totally unrelated since I can't access the website, but one thing that I'd like to see in a typing system used in data analysis is to be able to see the dimension of the data structures used. For instance, to be able to tell at compile time if a matrix multiplication is going to crash due to dimension mismatch.

So far, in Python at least, at best I can tell that it's a float array, without specifying the dimension (for example 2d for a matrix) or better, specifying its dimension is (n,p) where n and p are both type variables.

Quandl newsletter for curated and handpicked articles about the intersection of finance and machine learning.

https://alternative-data-news.quandl.com

Kinda did, but mostly the first chapters, actually up to CNN chapter (where real modern DL start). But so far, I really liked what I read. It has a very good blend of code and theory, with hands on applications throughout the whole book. Most importantly, all those applications could perfectly be copy pasted into your own environment. So it actually reminded me of a very thorough tutorial on a framework, more say than a regular textbook, although the authors don't compromise on mathematical arguments (but don't get lost in it either, they skimmed pretty fast on regularization theory imho). If you've had previous exposure to classical ML, I think it's a fantastic introduction to DL, enough to get started.

Python is like democracy. It's the worst programming languages except for all the others.

LaTeX is still one of my favourite piece of technology of all time. It is at time so alien, yet beautiful.

I now get closer to 10+ years of programming experience, yet nothing comes close to debugging a faulty LaTeX custom command... it can quickly turn to an unreadable mess, but you have to admit that once everything is swept under a preamble.tex file, the rest of the code is very clean. Especially with auctex in emacs which displays most math symbols as their true unicode counterpart.

Funny story: one of my first gig was working in a music instruments shop where I was basically the IT guy, from sysadmin to web dev. At some point the software that created the barcode labels stopped working. Now I had to find an automatic way to make those labels, so of course I turned to LaTeX. All I needed to do was to write a batch file calling `pdflatex` with a template tex file and a pdf file for the label was promptly sent to the printer! There is probably some python package for doing the same thing, but I was so proud of seeing Computer Modern font tagged to every instruments in the shop!

Erik Meijer a very engaging person! If I recall correctly, he spearheaded the Linq project for C# back in mid-2000s.

In particular, I suggest his Haskell lectures to anyone interesting by theoretical aspects of functional programming. https://channel9.msdn.com/Series/C9-Lectures-Erik-Meijer-Fun...

Interesting. Out of curiosity, why then did you chose France over any other EU country to have your company?

HN will likely rejoice. Macron is putting a lot of faith in startups to jumpstart back the french economy. I think it's wise: France has great schools to train great engineers, but there's still this stigma that it's almost impossible to have your own tech company in France unless you're ready to deal with massive paperwork. I have no idea how true it is, but Macron clearly wants to change this mentality.

It also goes to another level. Traditionally (since the 50s), engineering in France has largely been lead by large, ambitious and state funded projects and organizations. Think for instance Concorde, TGV, EDF, INRIA, etc. I have always liked this gaullist mentality: it has benefited the majority of the population and I tend to believe it has been helpful in making France a major post WW2 superpower.

But lately it may not have worked so well: for instance, many regional airports have been built at high cost, and yet many of them are almost empty. Perhaps then, resource allocation is the problem. In this regard, Macron is influenced by the american model where eg. Bell Labs, Elon Musk, Silicon Valley behemoths, etc. decide by themselves what shall be built and the state has little to no regard in the orientations of these organizations.

We'll see. I'm glad Macron has won and hope his plan works out.

Polls close at 7pm except in larger cities, but counting of votes start at this time. It's forbidden to leak the results until 8pm, at which time the sample is large enough to give a precise estimate, unlikely to move by much.

For those on MacOS, I suggest this port of emacs : https://bitbucket.org/mituharu/emacs-mac/

It's basically the same as the distribution from http://emacsformacosx.com, but it supports various enhancements made for mac, eg. resize text size with trackpad, smooth buffer scrolling and SVG support, which is quite convenient when used with the jupyter notebook interface and producing plots. See for instance : http://imgur.com/gallery/vEI2z.

Also consider Foundations of Machine Learning (2012) by Meryar Mohri. It is a very good book with a statistical learning approach to ML. If you're interested by statistical guarantees to ML algorithms, eg. why the SVM algorithm works and its sample complexity (the expected out of sample error in function of the sample size), it's the book you need.

Actually, there's a slight difference between the two pronunciations, at least for the North American English pronunciation. When you say cock, there's an emphasis on the o (it lasts longer) while with coq, the o is slightly muttered. Like c-awk vs. c-uck, or according to wikitionary: \kɒk\ vs. \kɔk\ for coq.

org is really slow when it gets to calculated column on a moderately sized table however.

Worth noting, Université Paris Diderot will launch its second edition of ocaml programming mooc in september: https://www.fun-mooc.fr/courses/parisdiderot/56002S02/sessio....

I followed along its first edition last year and it was very good. Sure, the instructors had a really thick french accent and the videos had some outdated sound effects and animations, but the concept were well presented. That said, the selling point of this mooc was the high quality of the exercises, which were rather hard, but quite rewarding and very well thought out. By week 2 or 3 you had to implement complex algorithms and functions, eg. a database class or heap's algorithm. Highly recommended.

Absolutely agree. Haven't read the english translation, but the original version was one of my favourite read last year. It is a short book (about 150pages I believe) and very skillfully written. It is very clinical, but in the same way Lovecraft is clinical in his descriptions of the the unnameable. It ends up giving a chilly insight on a deeply tormented man, who had trouble adjusting to modern society in all its horror and prurience!

Just saying, but if you want to hop onto the ML bandwagon (for instance), then don't bother going over linear algebra or probabilities first, and instead just learn what you need as you go. For example, the first sections of this book are already devoted to getting you on the right track, and it's somewhat standard to do so. And besides, there's no need in learning what are rotation matrices if you won't use them.

Using concentration inequalities on Lipschitz convex learning algorithms to derive generalizing bounds. The seminal papers for this would be Stability and Generalization by Bousquet and Elisseef (2002), or those by Shalev Schwartz.

That's nice to hear! I'm doing my masters thesis on statistical learning and I often think how un-glamour this field now is. No bayesianism, less engineering, but having probabilistic guarantees on your out of sample results as well as sample complexity, no matter the underlying distribution, can be quite beneficial.

I'm currently enrolled in a MSc program with a thesis on portfolio optimization using statistical learning. It might even lead to a publication! But now, the question is, should I go onto a PhD in stats or CS or just settle with it. I know I probably have the chops to complete it, it's just that 4 or 5 five more years in school is somewhat risky. Sometimes it's hard to assert your worth on such a competitive market.

On a side note: omg janestreet is such my dream job. Writing functional code, working on immensely interesting problems (what is the market distribution?) and being paid for it.

This reminds me of those educational/cultural interactive cdroms that were a thing in the 90s/early 2000s and that my local library had a for a 1-week rent. Most of them were kinda fun (for a kid) even though I wished in those times they had a copy of Diablo 1 instead! Anyway, it just seems to me that technology has lost the drive to produce this kind of content which was more prevalent before.

I'm also building a data driven (mostly NLP) portfolio optimizer, based on expected utility theory! It's a naive model, but it also has statistical bounds on its efficiency relative to the regret, which is an interesting bonus.

Out of curiosity, what kind of model (loss function) are you using?