From: <jc...@fe...> - 2003-09-26 14:38:02
|
On Friday 26 September 2003 05:56, Alan W. Irwin wrote: | On 2003-09-26 00:38+0100 Jo=E3o Cardoso wrote: | > On Thursday 25 September 2003 23:36, Pavel Sakov wrote: | > | Hi Folks, | > | | > | I probably should have been more explicit in my message to Joao. | > | | > | I checked svd.f from EISPACK, and this file is equivalent to | > | svdcmp() from NR. I mean, it is the source of the svd code in NR. | > | The NR authors did not even bother to rename the variables. | > | > Shameful! | | Actually not. They gave full credit in the NR book to prior work, | and that is all that is required. That prior work was all free so NR | could make small changes to the fortran source, convert to C and | relicense the code in any way they chose. | | To avoid that bad license, we just have to steer completely clear of | both the fortran and C NR versions and go back and rederive what we | need directly from EISPACK fortran source using f2c. I'm pretty satisfied with Pavel's approach to solve the problem, as the=20 following highlights from private e-mails show. When I will sync PLplot=20 with Pavel's csa everything will be OK: ... pavel > However, as you suggested, one can use the original code from=20 EISPACK, and that is what I have done. ... pavel > Compared with the previous version of SVD code in csa, most=20 of the changes were pretty cosmetic.=20 ... pavel > The only thing that arguably was borrowed from NR was the=20 idea of rewriting the input matrix A by output matrix U. [jc: =20 but this is normal practice in other fortran routines, if I=20 remember correctly] ... jc > But you will change the NR credits in the source file, asserting = =20 the facts, right? pavel > After syncing the code with EISPACK, I changed the credits to=20 EISPACK. ... pavel > Correction -- there is a reference to EISPACK Guide in the=20 SVD chapter. Does everybody agree? Joao |