APL in R | Hacker News Reader

APL in R

bookdown.org

84 pointsghosthamlet24 comments

The APL keyboard is the reason R's assignment operator, the classic one at least, is <-

Actually, I think that while there might have been some inspiration from APL, the basic reason is that ASR33 teletype machines have a backarrow character, whose ASCII code is the one more commonly associated with underscore. That explains why for many years the underscore character was accepted as an assignment operator in R and S. The <- glyph is a synonym useful if for some bizarre reason you are not programming with an ASR33 teletype machine.

That's an interesting point and likely correct.

The story I gave here was told to me way back in the day by RG himself, although the real reason R uses `<-` is "because that's what S does" and it's possible that the reason he was given wasn't 100% correct itself.

FWIW, the left arrow is also what smalltalk used for assignment, from wikipedia[0]

    In the original Parc Place image, the glyph of the underscore character (_) 
    appeared as a left-facing arrow (like in the 1963 version of the ASCII code). 
    Smalltalk originally accepted this left-arrow as the only assignment operator.

[0] https://en.wikipedia.org/wiki/Smalltalk#Assignment

the basic reason is that ASR33 teletype machines have a backarrow character, whose ASCII code is the one more commonly associated with underscore.

When I read this I said "huh?" because I used ASR33's quite a bit in my youth and I didn't remember that.

But, sure enough, the 1963 version of ASCII had the backarrow. It was replaced by underscore in the 1965 version. The backarrow would have been current when Iverson first developed APL.

https://en.wikipedia.org/wiki/ASCII#Printable_characters

APL is a fascinating language -- it keeps popping up in odd corners like this, and there are echoes of APL in the design of a lot of popular libraries. Numpy is a good example: it's almost another language embedded in Python.

agree with you, apl is great for ML, see this 6 lines Neural Network Example by apl with no deps: https://github.com/ghosthamlet/ann.apl

all code here:

      l0 ← 4 3 ⍴ 0 0 1  0 1 1  1 0 1  1 1 1

      y ← 4 1 ⍴ 0 0 1 1

      syn0 ← 3 1 ⍴ 0.5069797399 0.5703818579 0.9265880619

      i ← 0

      e ← 2.718281828459045

      :While i < 10000 ⋄ i ← i + 1 ⋄ d ← l0 +.× syn0 ⋄ l1 ← 1 ÷ 1 + e * 0 - d ⋄ l1_err ← y - l1 ⋄ l1_delta ← l1_err × l1 × 1 - l1 ⋄ syn0 ← syn0 + (⍉ l0) +.× l1_delta ⋄ :EndWhile

      l1

by the way, it is a ports from 11 lines python version here: http://iamtrask.github.io/2015/07/12/basic-python-network/, isn't it more readable concise and beautiful than python version?

if our all ML codes use apl, it will be a cool world.

> isn't it more readable concise and beautiful than python version?

No.

To add to this, the which-loop-in-one-line-of-code is not self-documenting, which is important for data/ML in particular when others need to verify your code.

And it's not any shorter in terms of statements, they're just using the diamond operator to put all the statements in the loop on one line.

It could simply be a badly written bit of APL. Although I don't know APL, I have some experience with J and I think the `:while` construct used here suggest more or less literal translation of the Python code.

Is it possible to put that while loop on multiple lines? It looks like there are 6 different assignments in there, and trying to pick out what is happening is very hard when it's all on one line like that.

You can break it up at each diamond but then you'd need to wrap it in a function which would be harder to cut and paste into a repl

Hmmmm. This is a bad example of APL. Source: I used APL professionally for over ten years. For one thing, real APL does not have constructs like "while". That's the whole point of the language, to remove the need for the programmer to have to think this way.

Also, this isn't, strictly speaking, six lines of code. The diamond is a statement separator. This is more like 15 or 20 lines (didn't count). Not that this is an important metric at all.

I'd be interested to know: for what kind of application did you use APL?

Quite a range, actually. From business applications to robotics and DNA sequencing.

gnu apl + emacs dyalog mode is a cute way to test

What font do you use for APL symbols?

I've used this on the Mac: http://www.dyalog.com/apl-font-keyboard.htm

Font and keyboard settings work independently of Dyalog APL. (Thank you, btw, for that!) For example, I've used them with GNU APL.

Just for the possible person using linux: the easiest way I've found to get that Dyalog font working is to install as it instructs you to, then run the bash script provided in this stackoverflow answer (http://stackoverflow.com/a/27955014/3430492).

emacs dyalog comes with a visual (textual) apl keyboard with the official symbols. There's also alternative bindings for more direct input.

Can someone provide a motivating example here? I'm not opposed to reading this, but don't understand why I should.

I wouldn't recommend reading the whole thing -- it's a manual. Read the introduction and maybe the "core" chapter. The intro is really short.

I don't have an example, but here are some motivations: Read it because you're generally interested in how people use programming languages to solve their problems. Read it because you use R and would like to have a really nice way of talking about array operations. Read it because you use APL and you want access to all of the neat stats functions R provides.

Reply on news.ycombinator.com