Re: AD matrix operations

[John W. Eaton]

On 01/26/2017 09:30 AM, Brad Bell wrote:
> The type a_double in Octave (and other languages)
>     http://www.seanet.com/~bradbell/cppad_swig/a_double_ctor.htm
> is a scalar type that will record floating point operations with the
> purpose of creating a corresponding function object
>     http://www.seanet.com/~bradbell/cppad_swig/a_fun_ctor.htm
>
> In python, you can create a numpy matrix of type a_double and perform
> operations, like matrix addition, multiplication, etc on the objects.
> The actual matrix loops are done in C (not in python). Can one do  a
> similar thing in Octave ?

If you are already creating a custom object that wraps a matrix, then 
you can just do something like this

  @a_double/mtimes.m
  function r = mtimes (x, y)
    ## compute the normal numeric result using Octave's builtin
    ## function for matrix multiplication.
    ## assumes "data" is the data member that contains
    ## the matrix that a_dobule wraps.
    numeric_result = builtin ('mtimes', x.data, y.data);
    ... do the other stuff that your a_double type needs to do ...
    ## package the result in an a_double object:
    r = a_double (...);
  endfunction

The above is the old-style class syntax, but you should also be able to 
do this with a classdef object and methods.  If there are problems, 
report the bugs.

jwe