For each training cycle the calculated and propagated error is recorded in
the error file, by convention <filename>.err.
This information is very useful to find out how fast a network is converging and in determining an optimal maximum training error. The format of the file is:
0 0.455532
1 0.426511
2 0.391984
...
999 0.001322
Files in this format can be read and visualized by many programs, including gnuplot on the UNIX system and MS-Excel on PCs.