On Sat, Mar 15, 2014 at 12:40 PM, Nathaniel Smith <njs@pobox.com> wrote:

On Sat, Mar 15, 2014 at 1:28 PM, Nathaniel Smith <njs@pobox.com> wrote:

On Sat, Mar 15, 2014 at 3:41 AM, Nathaniel Smith <njs@pobox.com> wrote:

Hi all,

Here's the main blocker for adding a matrix multiply operator '@' to Python: we need to decide what we think its precedence and associativity

should

be.

Another data point that might be useful:

Matlab: same-left

R: tight-left

I was going to ask this earlier, but I was worried I was missing something major.

Why was "tight-left" not an option?

This means that if you don't use parentheses, you get: a @ b @ c -> (a @ b) @ c a * b @ c -> a * (b @ c) a @ b * c -> (a @ b) * c

In my (very inexperienced) opinion, it seems like the most intuitive

On Sat, Mar 15, 2014 at 6:33 PM, Joe Kington <joferkington@gmail.com> wrote: option.

Because tight-left doesn't seem to have much to recommend it over same-left, and all else being equal having fewer levels of precedence is usually considered a good thing. Unless I'm missing something. If we do decide that tight-left is best then we could certainly advocate for it.

I wouldn't read too much into R's choice; they don't actually define a separate precedence level for matrix multiplication specifically. They have a single precedence level for all "special" (user-defined) operators, and matrix multiplication happens to be one of these. (Their versions of // and % are also "special", but I don't think anyone would expect // to bind more tightly than / if one were choosing precedences on a case-by-case basis.)

Just to throw something new into the mix u@v@w = u@(v@w) -- u@v is a dyadic matrix u@v -- is a scalar It would be nice if u@v@None, or some such, would evaluate as a dyad. Or else we will still need the concept of row and column 1-D matrices. I still think v.T should set a flag so that one can distinguish u@v.T (dyad) from u.T@v (inner product), where 1-D arrays are normally treated as column vectors. Chuck