OT: excellent book on information theory

[Anton Vredegoor]

Terry Hancock wrote:

> On 19 Jan 2006 13:57:06 +0100
> Anton Vredegoor <anton.vredegoor at gmail.com> wrote:
> > Some time ago I tried to 'sell' Python to a mathematician.
> > The crucial point was that it was not (in standard Python)
> > possible to have a matrix A and a matrix B and then do for
> > example:
> >
> > A = A * B
> >
> > and have a matrix multiplication performed.
>
> Um, why not? I'm trying to think what would be the stumbling
> block. I don't use matrix multiplies much, but I have
> implemented 3D vector math so that "*" is the "dot product"
> and "%" is the "cross product", which is pretty trivial to
> do.

Of course ! And to think that I even have used this trick a few times,
for example to implement set operations using long integers. I feel
ashamed.

In my defense I can only explain what happened. Four years ago (when I
knew a lot less of Python) I tried to use Numeric to do A*B for
matrices, but that resulted in something else than expected.

So I redefined the star operator by subclassing a *numeric python*
object but then it didn't work (the subclassing IIRC). Then it turned
out there was a Matrix module for Numeric that did exacly what was
asked, but by that time I was trying to understand Numeric by reading
the docs and 'selling Python' at the same time, which didn't work too
well ...

The main reason for that was that it was necessary to convince someone
not having any Python knowledge to install Python *and* some module
that I didn't know about and then that module needed *another* install
which I didn't know about, and the docs for Numeric were separate from
Python. Just too much at once.

I believe if I just had implemented matrix multiplication myself at the
time in plain Python I wouldn't have overcomplicated the matter in such
a way that I couldn't convince anyone else anymore :-)

So I got lost in Numerics complexities and that made me forget the
basic option.

By now I have used Numeric enough to make it likely that I could
explain its use to someone.

But even when I cured myself of this deficiency, the memory of failure
stayed in my head.

Witness a classic freudian fixation phenomenon in a Python learning
curve :-)

In order to prevent such mental damage for future Python programmers, I
propose to add a simple matrix multiplication module to the standard
distribution.

>
> The only obstacle I've run into is that you can't
> (easily) define *new* operators and precedence levels.
>
> There *is* a trick for doing this that was posted on
> the list some time back, which involved overloading an
> operator to "apply an operator":
>
> It would've allowed you to do something like this:
>
> a |dot| b
> a |cross| b
>
> or perhaps
>
> a <dot> b
> a <cross> b
>
> I don't remember where this is posted. The trick was in
> overloading the <, >, or | to interact specially with
> "operator" objects.

That's very nice. Thanks to you for mentioning this and to Jorge, who
provided the link to activestate for this recipe in another message.

Anton