Re: linalg[leastsqrs] in Maple V R4

lcw1964 wrote:
I have decided to write my own leastsqrs routine using linalg(Svd)
which seems to be a bit more consistent. This is how I have tested it
out and have convinced myself it is adequate for my own purposes.


There is one disadvantage to my approach evalf(Svd()) is palpably slow
for big matrices--I just tested the routine out with u being of
dimensions 136x17--and Digits set to something high like 40 or 50. That
said, the significant digit loss when I reconstitute the original u
matrix from the decomposed parts is still in the 3 digit max range. In
comparison, linalg[leastsqrs] spits out a result quickly, but I must
admit I don't trust it as much.

Les

.