Re: ublas & custom containers?

From: jhrwalter (walter_at_[hidden])
Date: 2002-03-14 05:14:11

--- In boost_at_y..., Kresimir Fresl <fresl_at_g...> wrote:
> 
> rwgk wrote:
> 
> [...]
> 
> > Unfortunately I am having trouble compiling this minimal
> > example:
> >
> > #include <vector>
> > #include <boost/numeric/ublas/config.h>
> > #include <boost/numeric/ublas/vector.h>
> > 
> > int main()
> > {
> >   numerics::vector<float, std::vector<float> > ua(1000);
> >   ua = ua * ua;
> >   return 0;
> > }
> > 
> > Attached are the minimal patches that I applied to config.h
> > and iterator.h, and the output from two compilers.
> > Advice on how to fix the problems is very much appreciated.
> 
> [...]
> 
> > gcc 3.0.4 (RedHat 7.1):
> > 
> > g++  -fPIC -ftemplate-depth-50 -DUSE_GCC  -g  -
> > I"/net/cci/rwgk/boost"  -c ublas_shared.cpp
> > ublas_shared.cpp: In function `int main()':
> > ublas_shared.cpp:8: ambiguous overload for 
`numerics::vector<float, 
> >    std::vector<float, std::allocator<float> > >& * 
> > numerics::vector<float, 
> >    std::vector<float, std::allocator<float> > >&' operator
> 
> [...]
> 
> It seems that error messages are misleading. There is
> no ambiguity in your example. Rather, in ublas there is no
> 
>      operator* (vector<> const&, vector<> const&)
> 
> or, more precisely,
> 
>      operator* (vector_expression<> const&,
>                        vector_expression<> const&)
> 
> My guess is that authors decided not to provide this operator
> because it is not clear whether it denotes inner product
> or elementwise multiplication.

This is also an issue of MSVC compatibility.
 
> `operator*()' can only be used to multiply vectors (and matrices)
> with scalars. (So, you can change the test operation to
> `ua = 2. * ua;'. Or, you can use `+' or `-'.)
> 
> To compute inner product you must use function `inner_prod()'.
> 
> As far as I know, there is currently no way to do elementwise
> multiplication. Strictly speaking, this is not linear algebra 
operation,
> but I hope that authors will in some future version add some means
> to do this and similar array operations (eg. apply some function to
> all elements in a vector).

I see two alternatives to extending ublas: element-wise operations 
could be a nice extension of boost.multi_array. Or one could use 
something like the expression templated valarray<> of gcc (but I'm 
unsure, if valarray<> is a legal array_type of numerics::vector 
currently ;-).
 
Regards
 
Joerg