Hygenic Macros

[David Pokorny]

Hi,

Just wondering if anyone has considered macros for Python. I have one
good use case. In "R", the statistical programming language, you can
multiply matrices with A %*% B (A*B corresponds to pointwise
multiplication). In Python, I have to type

import Numeric
matrixmultiply(A,B)

which makes my code almost unreadable.

Thanks,
David

 

[Robert Kern]

Hi,

Just wondering if anyone has considered macros for Python. I have one
good use case. In "R", the statistical programming language, you can
multiply matrices with A %*% B (A*B corresponds to pointwise
multiplication). In Python, I have to type

import Numeric
matrixmultiply(A,B)

which makes my code almost unreadable.

Well, dot(A, B) is better. But if you must:

http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/384122

--
Robert Kern
(e-mail address removed)

"In the fields of hell where the grass grows high
Are the graves of dreams allowed to die."
-- Richard Harter

 

[Dan Farina]

Hi,

Just wondering if anyone has considered macros for Python. I have one
good use case. In "R", the statistical programming language, you can
multiply matrices with A %*% B (A*B corresponds to pointwise
multiplication). In Python, I have to type

import Numeric
matrixmultiply(A,B)

which makes my code almost unreadable.

Thanks,
David

The problem here is that Python's parse trees are of non-trivial ugliness.

A page on the compiler.ast module:
http://docs.python.org/lib/node792.html

it is, in fact, perfectly possible to write yourself a pre-processor for
your particular application. You may have to fiddle with the token you
want for notation depending on how the AST fleshes out (% is used by at
least a couple of things, after all). My cursory familiarity with
python grammar suggests to me that this particular choice of token could
be a problem.

I would say try it and see. Keep in mind though that since Python's AST
is not a trivial matter like it is in Lisp and the like that doing
metaprogramming of this sort probably falls into the category of black
magic unless it turns out to be very trivial.

Another option is to define your own tiny class that will override the
__mult__ method so that you can simply do:

A * B

Which may not be what you want.

df

 

[Steven D'Aprano]

Hi,

Just wondering if anyone has considered macros for Python. I have one
good use case. In "R", the statistical programming language, you can
multiply matrices with A %*% B (A*B corresponds to pointwise
multiplication). In Python, I have to type

import Numeric
matrixmultiply(A,B)

which makes my code almost unreadable.

Yes, I see what you mean, it is pretty confusing. It almost looks like
a function that multiplies two matrices and returns the result.

Have you tried coming up with better names for your arguments than A and
B? Many people find that using self-documenting variable names helps make
code easier to understand.

 

[Robert Kern]

Yes, I see what you mean, it is pretty confusing. It almost looks like
a function that multiplies two matrices and returns the result.

Have you tried coming up with better names for your arguments than A and
B? Many people find that using self-documenting variable names helps make
code easier to understand.

Well, to be fair, his example was trivial. When you have more
complicated matrix expressions with transposes and conjugations and more
matrixmultiplies than you can shake a stick at, it gets ugly pretty fast.

F = dot(dot(Z, F),transpose(conjugate(Z)))

versus

from scipy import *
F = mat(F)
Z = mat(Z)
F = Z*F*Z.H

--
Robert Kern
(e-mail address removed)

"In the fields of hell where the grass grows high
Are the graves of dreams allowed to die."
-- Richard Harter

 

[Steven D'Aprano]

Well, to be fair, his example was trivial. When you have more
complicated matrix expressions with transposes and conjugations and more
matrixmultiplies than you can shake a stick at, it gets ugly pretty fast.

F = dot(dot(Z, F),transpose(conjugate(Z)))

No uglier than y = sin(poly(sqrt(x)))

And of course it is allowed to do this:

F = dot(Z, F)
Z = transpose(conjugate(Z))
F = dot(F, Z)

Not everything has to be a one-liner, not even in mathematics.

versus

from scipy import *
F = mat(F)
Z = mat(Z)
F = Z*F*Z.H

It's a matter of taste. But calling it "almost unreadable" is an
exaggeration.

 

[David Pokorny]

Hi,

Thanks - this cookbook entry is very cool!

I am somewhat worried about function call overhead from the infix hack
though... on second consideration, the infix issue is not as important
as eventually boosting the speed of the inner loop to which
matrixmultiply() belongs. For those who are interested in things like
fast math, I found

http://www.scipy.org/documentation/weave/weaveperformance.html

very eye-opening. For many of the examples cited there, it is possible
to simply treat the source as a string and then do the necessary macro
transformations before passing it off to a compiler...

David