Next: 'stopper', Maximum Squared Error
Up: Base Modules
Previous: sigmoidalUnit.m, sigmoidalUnitDeriv.m, valueUnit.m, valueUnit2Deriv.m,
ConvChk.m and ConvChkGlob.m
Module that checks for convergence. Returns true (1) if converged, false (0)
otherwise.
The variables used in this module are as follows.
- sqErr
- The square of the matrix benefL2
(see Section 4.1.9).
- MSE
- The scalar value which is the mean value of all the
elements in the matrix sqErr .
- msErr
- The array holding MSE values
for each pSS (see Section 4.1.1) number
of epochs, therefore an interpretation of the progress in time of the training
session. This variable is plotted across number of epochs at the end of training
(see Section 3.1.3.1).
- HitTally
- Number of elements of the matrix sqErr
that are below the given error criterion, i.e. acceptable outputs of units in
all patterns.
- rejected
- Used as convergence criterion in classifier
networks with 1-out-of-many encoding at the output layer, where only one unit
should be active at any one time. It holds the number of rejected patterns where
there are more than one output unit active and close to each other more than
the value of the rejectionCriterion (see
Section 4.1.1) variable.
- stopper
- Maximum element of the matrix sqErr ,
therefore the one largest squared error value of the all output unit values
across all patterns, i.e. maximum error.
- Misses
- Inverse meaning of the variable HitTally ,
therefore number of elements of the matrix sqErr that
are above given error criterion, therefore outputs across all patterns that
are not acceptable.
- allstopper
- The array holding stopper
values for each pSS (see Section 4.1.1)
number of epochs, therefore an interpretation of the progress in time of the
training session. This variable is plotted across number of epochs at the end
of training (see Section 3.1.3.1).
After calculating the variables needed for the convergence check, a message
is printed out giving information about all the current error values. Then,
according to the selected error criterion a switch statement is executed and
the convergence check is decided and returned as a boolean value from the function.
The possible convergence check criteria are described in the following subsections,
according to the value of the variable checkType
defined in Section 4.1.1.
Next: 'stopper', Maximum Squared Error
Up: Base Modules
Previous: sigmoidalUnit.m, sigmoidalUnitDeriv.m, valueUnit.m, valueUnit2Deriv.m,
Cengiz Gunay
2000-06-25