The toolbox is too new to have a section with frequently answered questions. Instead we have this FEQ, which is a placeholder for answers that we do not want to be bugged about.
This is normal; large-scale computations can be long winded. The primary motivation for creating the toolbox was to make certain large-scale computations possible; speed was a secondary goal. We rely on Moore's law for speed.
With this said, the underlying algorithms in the toolbox do a decent job at exploiting modern hardware. The implementation itself, however, does not suffer from the "premature optimizations syndrome" yet, and it will be quite a long time before this is likely to happen. Implementing the toolbox mostly in MATLAB is good for development, but it does add some overhead.
The toolbox allows the user to make a trade-off between memory consumption and computational effort. As long as you have enough memory you should be able to use default parameters without further ado. However, tweaking the computation to a problem you know intimately may not just save memory but also speed up the computation.
This is normal; the author admits he is insane.
This is normal; when you compute N
of the largest
singular triplets/values, there will not generally be any guarantee that it will
be the N
largest singular triplets. The cure is to
compute a few extra triplets and discard any unwanted triplets.
NOTE: The documentation has not been completed yet. Expect a number of upgrades in the near future.
The documentation is primarily intended to be viewed inside Matlab, using the help-browser shipped with Matlab. The documentation published on the Web is an extra service. Frames are used to approximate the interface seen in Matlab.
Give the index-page an argument with the page you want to address, e.g.:
<a href="http://nru.dk/bbtools/help/toolbox/bbtools/index.html?feq.html">FEQ</a>
NOTE:The link above is not guaranteed to be stable. Furthermore, it will only work for clients with JavaScript enabled.