Hi,

the use of a custom function for .unaryExpr() is a bit clumsy.
Using the CwiseUnaryOp struct is probably the most versatile approach, but in most cases the user just wants to specify a specific function Scalar func(Scalar).

Thus I wrote this piece of code:

Code: Select all

  template<typename Scalar, Scalar Func(Scalar)>
  struct CwiseOp {
    const Scalar operator()(const Scalar& x) const { return Func(x); }
  };


Though, it would be great, if one could use .unaryExpr<func_name>() additionally to .unaryExpr(CwiseUnaryOp).
So .unaryExpr<func_name>() would just call .unaryExpr( CwiseOp<Scalar, func_name> ).

The only problem with my struct is, that you have to declare it for any return and argument type:

Code: Select all

Scalar Func(Scalar)
Scalar Func(const Scalar)
Scalar Func(const Scalar&)
Scalar Func(Scalar &)
const Scalar Func(Scalar)
const Scalar& Func(Scalar&)
const Scalar& Func(const Scalar&)
const Scalar Func(const Scalar)
const Scalar& Func(const Scalar&)
...


What do think about that?

Hi,

you can already do that using STL's ptr_fun construct, e.g.:

B = A.unaryExpr(std::ptr_fun(fabs));

oh, ok. Thanks!

Would be great if it would be added to the documentation of unaryExpr(), because I guess this is a common problem.


Cheers,

Manuel

Hi, it's me again!

Is there any possibility to execute the unaryExpr() only on non-zero elements of a sparse matrix?

Cheers,

Manuel

this is already the case.

Oh, ok. That was an easy one!