On Sun, Mar 21, 2010 at 8:01 AM, Ralf Gommers <[email protected]>wrote:
> > > On Sun, Mar 21, 2010 at 9:58 PM, Ralf Gommers <[email protected] > > wrote: > >> >> >> On Sun, Mar 21, 2010 at 9:51 PM, Alan G Isaac <[email protected]>wrote: >> >>> On 3/21/2010 12:54 AM, Ralf Gommers wrote: >>> > too many blank lines are needed >>> >>> Please define "need" after seeing the compact example I posted. >>> >>> You need 4 blank lines in your example. Now I tried adding a description >> for the first argument (q) like this: >> >> q, r if mode = 'full' : >> - q : ndarray of float or complex, shape (M, K) >> Description of `q`. >> >> - r : ndarray of float or complex, shape (K, N) >> >> That doesn't work, you need yet more blank lines (try this in the wiki >> editor). >> >> >> I just changed the docstring to the following, looks much better in both >> plain text and html imho: >> >> >> q : ndarray of float or complex, optional >> The orthonormal matrix, of shape (M, K). Only returned if >> ``mode='full'``. >> r : ndarray of float or complex, optional >> The upper-triangular matrix, of shape (K, N) with K = min(M, N). >> Only returned when ``mode='full'`` or ``mode='r'``. >> a2 : ndarray of float or complex, optional >> Array of shape (M, N), only returned when ``mode='economic``'. >> The diagonal and the upper triangle of `a2` contains `r`, while >> the rest of the matrix is undefined. >> > > This line in the code is fairly amusing by the way: > # economic mode. Isn't actually economic. > > Economic mode is very similar to 'r' mode anyway, what's the point? > > Economic mode is what the low level algorithm likely returns, it contains the info needed to contruct q if needed, or to efficiently apply q to different vectors without constructing q; constructing q adds to the computational and memory costs, as does pulling r out of the economic return. The situation is analogous to the LU decomposition where the natural form is to store both L and U in the original matrix. Other algorithms can then use that compact form to solve equations with different right hand sides. Chuck
_______________________________________________ NumPy-Discussion mailing list [email protected] http://mail.scipy.org/mailman/listinfo/numpy-discussion
