-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Just an idea for algorithm improvment(?) I came up with:
Is it possible, when entering the Evaluation function, to pivot the weight matrix and the input vector and store the order in a local index array. We sort the input vector (and the weight matrix) such that all the zeros are at the the end.
When the multiplication then reaches the first zero, it's breaking out of the loop. Simplifies the main loop by several flops.
Then we sort ar back the right order according to the saved index array.
This was just an idea I had .... is it smart?
How many input zeros are there in an average evaluation? and how many average zeros are there in average evaluation-from-base?
- -�ystein -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (MingW32) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
iD4DBQFCZqey6kDTFPhwyqYRAh7MAJ9sAeVJ/4l96uWvTz5SWjT8WG9EwwCYvhi4 8UT8+kSAd1EsQxHAh2SG1A== =Cppc -----END PGP SIGNATURE-----
_______________________________________________ Bug-gnubg mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gnubg
