[NumPy-Tickets] [NumPy] #1402: Provide a function to compute the log-determinant
NumPy Trac
numpy-tickets@scipy....
Fri Apr 30 14:22:19 CDT 2010
#1402: Provide a function to compute the log-determinant
--------------------------+-------------------------------------------------
Reporter: njs | Owner: somebody
Type: enhancement | Status: needs_review
Priority: normal | Milestone:
Component: numpy.linalg | Version:
Keywords: |
--------------------------+-------------------------------------------------
Comment(by njs):
Oh, duh, I'm an idiot -- you mean that the 'log(absd, absd)' line in the
implementation is confusing! Well, I'm not sure. Certainly a comment
wouldn't hurt, but as a core linear algebra routine, this seems to me
exactly the sort of place where we should err on the side of speed over
readability. It's probably not a big hit -- the array in question is the
diagonal of the matrix whose determinant is being taken, so its size is
O(sqrt(n)), and we already have to copy it once a few lines earlier (absd
= absolute(d)). But the cost *is* entirely superfluous.
On the other hand, that earlier copy could be eliminated (or its cost
reduced, at least) for real inputs by using np.sign, but it didn't seem
worth it to write separate branches for real and complex inputs, so I
can't exactly claim to be consistent.
--
Ticket URL: <http://projects.scipy.org/numpy/ticket/1402#comment:8>
NumPy <http://projects.scipy.org/numpy>
My example project
More information about the NumPy-Tickets
mailing list