More on this: (the original problem, recall, was that J's matrix inversion
failed to detect that a floating-point y'y matrix was singular when y
had two identical columns of data).
I tried the same experiment in APL-PLUS-Win 1.2 (circa 1990).
Quad-divide correctly decided that the matrix was singular.
In J, I rounded the original matrix to 2 decimal places and tried again.
Without x: the system did not recognize the singularity; with x: it did.
Which raises the question: Are there any settings with floating-point
numbers where for matrix inversion one would NOT want to use x: ? If
not, perhaps the system should do this automatically? And are there
other built-ins subject to similar recommendations?
------------------------
Philip A. Viton
City Planning, Ohio State University
275 West Woodruff Avenue, Columbus OH 43210
[EMAIL PROTECTED]
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
