Re: [eigen] documentation: the long tutorial

[ Thread Index | Date Index | More Archives ]

2010/6/21 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
> 2010/6/20 Carlos Becker <carlosbecker@xxxxxxxxx>:
>> Some new stuff:
>> Arithmetic
>> ops:
> Thanks. My main comment is that this needs a better structuring. For
> example, the first example already shows the comma-initializer without
> an explanation and already shows various products. I think the
> comma-initializer deserves a short paragraph for itself (which is OK
> to put in this page) and products totally deserve a paragraph for
> themselves before you do any (except scalar-matrix product, that is OK
> to do without a former explanation).
> It's good to explain right away that Matrix is only for linear algebra
> and to forward-link to Array for coeff-wise ops. But rowwise and
> colwise addition and substraction are actually available for matrices
> as well (rowwise() is in DenseBase, not in ArrayBase) so that's not
> the right example ;) since you just mentioned scalar+array, stick to
> that example.
> In the paragraph "Combining operators in a single statement" you give
> an example of how that allows to avoid useless temporaries, but in
> that example, actually, that temporary is not useless at all (write
> down the computation yourself on pen&paper, for
>    tempMatrix = m + n/2;
>    v = tempMatrix * v;
> you'll see that doing the temporary actually reduces complexity) and
> Eigen is clever enough to understand that, so actually in the 2nd
> version in one line, Eigen internally uses such a temporary.

Sorry! I read too fast. Indeed, since your right hand side is a
vector, the temporary is actually useless. But the fact that this
depends on v being a vector rather than a matrix makes it a rather
subtle example - i would still prefer an example as described below,
if you want to write a paragraph about that.

> If you want a good example of combining arithmetic operations into a
> big expression to avoid temporaries, the easiest way to get a good
> example is to just use additions and scalar-matrix multiplications.
> But I would not mention that at the top of this page, I'd rather put
> at the end of it, since it's an advanced topic.
> In short: we need this page to expose all the basic arithmetic in a
> rational, systematic, and progressive way, treating one thing at a
> time, and deferring the advanced topics to the end.
> Also, regarding your first FIXME, you can remove for now the
> navigation links at the top.
> Benoit
>> Array class:
>> The Array class part is quite incomplete I guess. I am wondering whether
>> more details should be added or if this is enough. I think that the examples
>> found in are quite useful,
>> does anyone know in which .dox file they are found?
>> All feedback welcome.
>> Carlos

Mail converted by MHonArc 2.6.19+