Forum: Ruby Bug in Vector class?

Announcement (2017-05-07): www.ruby-forum.com is now read-only since I unfortunately do not have the time to support and maintain the forum any more. Please see rubyonrails.org/community and ruby-lang.org/en/community for other Rails- und Ruby-related community platforms.
Dave B. (Guest)
on 2008-10-09 13:54
(Received via mailing list)
require 'matrix'
m = Matrix.identity(4)
v = Vector[5, 6, 7, 1]
puts m * v

gives the expected answer of [5, 6, 7, 1]

but
puts v * m

gives
ExceptionForMatrix::ErrDimensionMismatch: Matrix dimension mismatch

because the Vector#* method promotes self to a column vector when the
argument is found to be a Matrix.  I think it should be promoting self
to a row vector and making this change removes the exception and gives
the correct result.  The updated method is:

  def *(x)
        case x
          when Numeric
              els = @elements.collect{|e| e * x}
              Vector.elements(els, false)
          when Matrix
              Matrix.row_vector(self) * x
          else
              s, x = x.coerce(self)
              s * x
          end
   end


Dave.
Thomas B. (Guest)
on 2008-10-09 19:15
Dave B. wrote:
> because the Vector#* method promotes self to a column vector when the
> argument is found to be a Matrix.  I think it should be promoting self
> to a row vector and making this change removes the exception and gives
> the correct result.

I think it should not. Either the programmer knows what he or she is
doing and is able to take care of the matrices dimensions, or no magic
auto-transpositions or other tricks are going to save their project from
falling apart sooner or later. Vector is a column vector, and let it be
this way.

TPR.
Dave B. (Guest)
on 2008-10-09 19:51
(Received via mailing list)
On 9 Oct 2008, at 10:40, Dave B. wrote:

> gives
>               els = @elements.collect{|e| e * x}
> Dave.
>

Further modification so it returns a Vector instead of a matrix
              (Matrix.row_vector(self) * x).row(0)

Dave.
Dave B. (Guest)
on 2008-10-09 21:03
(Received via mailing list)
On 9 Oct 2008, at 16:13, Thomas B. wrote:

> auto-transpositions or other tricks are going to save their project
> from
> falling apart sooner or later. Vector is a column vector, and let it
> be
> this way.


The Vector class doesn't seem to have any idea that it is representing
a row or column vector and as the class stands now a vector * matrix
will always fail.

The only way to achieve this operation is to manually convert the
vector into a Matrix with one row but now you are really doing a
matrix * matrix operation and not using the failing path in Vector.

If the intent was for vector * matrix to fail then it would be better
to raise an exception directly rather than rely on the Matrix class to
catch a dimensional error.

I still think it is better to treat it as a row vector in this case.

Dave.
Thomas B. (Guest)
on 2008-10-09 21:14
Dave B. wrote:
> The Vector class doesn't seem to have any idea that it is representing
> a row or column vector and as the class stands now a vector * matrix
> will always fail.

Vector.elements([1,2,3])*Matrix.identity(1)
#=> Matrix[[1], [2], [3]]

Of course it's not much of a useful operation, but for me it's evident
from it that a vector assumes itself to be a vertical matrix each time
it is required to behave like a matrix.

And now, what will be the result of the above operation if, as you want,
a vector would auto-transpose when sent * with a matrix? A row or a
column matrix?  For me, this proves that your idea is inconsistent. It
is not true that vector*matrix operation is never possible, so we
shouldn't automagically make it possible even if it is not. It wouldn't
make debugging any easier.

TPR.
Dave B. (Guest)
on 2008-10-10 11:43
(Received via mailing list)
On 9 Oct 2008, at 18:12, Thomas B. wrote:

> from it that a vector assumes itself to be a vertical matrix each time
> it is required to behave like a matrix.
>
> And now, what will be the result of the above operation if, as you
> want,
> a vector would auto-transpose when sent * with a matrix? A row or a
> column matrix?  For me, this proves that your idea is inconsistent. It
> is not true that vector*matrix operation is never possible, so we
> shouldn't automagically make it possible even if it is not. It
> wouldn't
> make debugging any easier.

OK, so you demonstrated a degenerate case that is not much use as you
point out.  How would you do something more useful such as
Vector[1,2,3]*Matrix.identity(3)?  I don't think you can unless you
convert the vector to a row vector aka a matrix with one row and do it
as a matrix * matrix operation.

Also a Vector * Matrix should return a Vector and not a Matrix (as the
Matrix * Vector operation does).

Thankfully with open classes I can get the behaviour I think is
desirable.

Dave.
This topic is locked and can not be replied to.