Talk:QR decomposition: Difference between revisions
Content added Content deleted
(→Questionnable intent: new section) |
|||
Line 69: | Line 69: | ||
Is this enough? Or should they implement the algorithm given in the description? |
Is this enough? Or should they implement the algorithm given in the description? |
||
I think they should show the algorithm implemented not just the function called! |
I think they should show the algorithm implemented not just the function called! |
||
== Questionnable intent == |
|||
The task is misleading at best. Some answers successfully compute Householder projections, |
|||
but, like the task description, they fail to understand that a QR decomposition is never |
|||
computed this way, as this would be too much time- and space- consumming. Only the vector |
|||
''u'' of array ''I- s uu''' is ever stored, together with ''s'' or something equivalent to ''s''. |
|||
And both ''Q'' and ''R'' are stored in ''A'' in the process, with only a supplementary vector. |
|||
LINPACK and LAPACK are of course no exception, though they handle the question in a [slightly |
|||
different way](http://stackoverflow.com/questions/3031215/mystified-by-qr-q-what-is-an-orthonormal-matrix-in-compact-form). |
|||
And when solving a system given an already computed QR in packed form, there are also ways to do |
|||
it in a clever way, not computing the ''Q'' and ''R'' matrices effectively. |
|||
Most answers only show bad ways to do the job, and they are just showing programming language |
|||
in (bad) action. |
|||
While I understand that Rosetta Code is not the place for state-of-the-art algorithms, it should be at least |
|||
mentionned that they are indeed very poor in this case. And anyway, a reasonnable QR algorithm is not more difficult |
|||
to implement, it just requires some work. |
|||
[[User:Arbautjc|Arbautjc]] ([[User talk:Arbautjc|talk]]) 11:05, 30 April 2015 (UTC) |