24-Feb-00

Hello,

there is (at least) one bug in MLPfit 1.40: weighting the examples
doesn't work properly with the "global" learning methods (BFGS,
Hybrid...) The bug is found and I will propagate the fix tomorrow
(Friday) to all libraries / executables. Please reload them tomorrow
evening from the web page if you need the fix.

In addition, Olivier Couet and myself are building some "test" paw
executables with MLPfit 1.40 (at least on Linux and alpha). Let me know
if you want the new PAW version before it goes to the official CERNLIB.
(I remind you that the new version is 40% faster than the previous one).

Experience shows that there is never enough testing...

       Jerome