Re: [eigen] Custom expression type API stability

[Dan Čermák <dan.cermak@xxxxxxxxxxxxxxxxxxx>]

On Thursday, February 18, 2016 06:57:42 PM Gael Guennebaud wrote:
> On Thu, Feb 18, 2016 at 1:46 PM, Dan Čermák <dan.cermak@xxxxxxxxxxxxxxxxxxx>
> wrote:
> > I need basically two different expressions:
> > 
> > 1. given two vectors:
> >         cond_vec and choice_vec -> result(i) = choice_vec(i) if
> > 
> > cond_vec(i) > 0
> > 
> > else choice_vec(i+1)
> > this is rather simple and maybe doable otherwise (and I have already
> > managed
> > that using the 3.3 development version, although it's probably ugly)
> 
> what about:
> 
> choice_vec.head(n-1) = (cond_vec>0).select(choice_vec.head(n-1) ,
> choice_vec.tail(n-1));

That is exactly what I need, thanks!

Maybe a small follow-up to that, is it possible to append/prepend a value to  
(cond_vec>0).select(choice_vec.head(n-1), choice_vec.tail(n-1)) ? I want to 
extend this vector of length n-1 to length n, but not create an additional 
vector along the way, so not something like this:
new_vec << 1, (cond_vec>0).select(choice_vec.head(n-1), choice_vec.tail(n-1))

Actually its a tad more complicated than that, since I want to take the choice 
result, put it into head(n-1) add subtract it from tail(n-1) and add values to 
the first and last element (while doing the same again with another 
expression).
Is there an elegant/efficient way how to do this? Or would it be best to give 
the function a reference to a result vector where to write the values into?

> 
> > 2. calculate reductions of 3d matrices times 2d matrices (they are not
> > stored
> > fully, since they can be calculated fairly easily from e.g. a matrix and a
> > vector, or are highly sparse): result(i) = Sum_j,k M_ijk * K_jk (as a full
> > reduction) or result(i,j) = Sum_k M_ikj * K_kj
> > I would like to create some abstraction where I only need to provide a
> > function which calculates the elements or row/col of the contraction
> > result.
> 
> Since M and K are not even Eigen matrices, then you can workaround with a
> nullary-functor, e.g.:
> 
> res = MatrixXd::NullaryExpr(rows,cols,your_fancy_functor(M,K));
> 
> where your_fancy_functor has to implement an operator()(Index i, Index j)
> method.
> 
> See:
> http://eigen.tuxfamily.org/dox-devel/classEigen_1_1DenseBase.html#a5a75198ed
> ac1ad62205648445ee30d34
> 
> And of course you can hide the construction of the nullary-expression using
> a free function, so that you simply has to call:
> 
> res = fancy_product(M,K);

This seems to be again exactly what I need, thanks for pointing me to the 
respective classes.


Cheers,

Dan

> 
> gael
> 
> > Cheers,
> > 
> > Dan
> > 
> > On Thursday 18 February 2016 12:10:07 Gael Guennebaud wrote:
> > > Indeed, writing custom expressions still requires to play with
> > > Eigen::internal, meaning that it is subject to change in the future. As
> > 
> > you
> > 
> > > noticed, the current 3.2 way is different than the future 3.3 way.. If
> > > you
> > > tell us a bit more about your need for a custom expression we might be
> > 
> > able
> > 
> > > to guide you towards the best approach. Maybe you don't need a new one
> > > at
> > > all, maybe ReturnByValue if good for, etc.
> > > 
> > > cheers,
> > > gael
> > > 
> > > On Thu, Feb 18, 2016 at 11:06 AM, Dan Čermák <
> > 
> > dan.cermak@xxxxxxxxxxxxxxxxxxx
> > 
> > > > wrote:
> > > > 
> > > > Hi Folks,
> > > > 
> > > > I have a question concerning adding custom expression types as
> > 
> > described
> > 
> > > > here:
> > > > http://eigen.tuxfamily.org/dox-devel/TopicNewExpressionType.html . I
> > 
> > am
> > 
> > > > currently developing a program that would require some custom
> > 
> > expression
> > 
> > > > types
> > > > to be efficient.
> > > > Is the way how they are implemented expected to change during the next
> > > > versions? Since I could not compile the example with Eigen 3.2.7 and
> > > > 3.2.8,
> > > > only with the latest snapshot from mercurial.
> > > > 
> > > > 
> > > > Cheers,
> > > > 
> > > > Dan