|
 TRAINING
Keyword of Integer
Type This keyword indicatea method that will be used to optimize
neural network weights. All these methods, with an exception of stiff and
Levenberg-Marquardt algorithms, are heuristic algorithms that have a number
of adjustable parameters. These parameters were selected by the authors
of the algorithms to provide the fast convergence of the algorithms.
 momentum {0}
-- The simplest algorithm that started new interest in artificial neural
networks after article of Rumelhart et al., 1986. The momentum rate is
0.9 and eps is 0.2.
SuperSAB {1}
-- programmed according to Tollenaere, T. SuperSAB: Fast Adaptive Back
Propagation with Good Scaling Properties. Neural Networks. 1990, 3, 561-573.
Momentum is 0.9, eps is 0.2, multiplication(+) is 1.05, multiplication(-)
is 0.5, max eps is 10.
RPROP {2} --
Initial eps is 0.0001, minimal eps is 0.000001, multiplication(+) is 1.2,
multiplication(-) is 0.5.
QuickProp {3}
-- Falhman implementation updated from his original code by V.V. Kovalishyn.
Eps is 0.55, decay is 0.0001, minimal derive is 0.1, maximal momentum is
1.75.
stiff {4}--
Second order optimisation algorithm that converts the weights optimisation
to the problem of solution of the second order differential equations.
QuickProp1 {5}
-- Original implementation of I. V. Tetko inspired by the algorithm of
Falhman. The same parameters as in QuickProp are used.
Marquardt {6}
-- Second -order Levenberg-Marquardt optimisation algorithm (requires O(n2)
time, where n is the number of weights), see e.g. Shepherd, A. J. Second-Order
Methods for Neural Networks; Springer-Verlag: London, 1997; 145.
The default value is {1}, the SuperSAB algorithm.
See FAQ if you have questions. How to cite this applet? Are you looking for a new job in chemoinformatics? |
|
