The function trains a deep architecture with the resilient backpropagation algorithm. It is able to use four different types of training (see details). For details of the resilient backpropagation algorithm see the references.
rpropagation(darch, trainData, targetData,
rprop.method = getParameter(".rprop.method"),
rprop.decFact = getParameter(".rprop.decFact"),
rprop.incFact = getParameter(".rprop.incFact"),
rprop.initDelta = getParameter(".rprop.initDelta"),
rprop.minDelta = getParameter(".rprop.minDelta"),
rprop.maxDelta = getParameter(".rprop.maxDelta"),
nesterovMomentum = getParameter(".darch.nesterovMomentum"),
dropout = getParameter(".darch.dropout"),
dropConnect = getParameter(".darch.dropout.dropConnect"),
errorFunction = getParameter(".darch.errorFunction"),
matMult = getParameter(".matMult"), debugMode = getParameter(".debug", F,
darch), ...)The deep architecture to train
The training data
The expected output for the training data
The method for the training. Default is "iRprop+"
Decreasing factor for the training. Default is 0.6.
Increasing factor for the training Default is 1.2.
Initialisation value for the update. Default is 0.0125.
Lower bound for step size. Default is 0.000001
Upper bound for step size. Default is 50
See darch.nesterovMomentum parameter of
darch.
See darch.dropout parameter of
darch.
See darch.dropout.dropConnect parameter of
darch.
See darch.errorFunction parameter of
darch.
Matrix multiplication function, internal parameter.
Whether debug mode is enabled, internal parameter.
Further parameters.
RPROP supports dropout and uses the weight update function as
defined via the darch.weightUpdateFunction parameter of
darch.
The code for the calculation of the weight change is a translation from the MATLAB code from the Rprop Optimization Toolbox implemented by R. Calandra (see References).
Copyright (c) 2011, Roberto Calandra. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. The names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. 4. If used in any scientific publications, the publication has to refer specifically to the work published on this webpage.
This software is provided by us "as is" and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for particular purpose are disclaimed. In no event shall the copyright holders or any contributor be liable for any direct, indirect, incidental, special, exemplary, or consequential damages however caused and on any theory of liability whether in contract, strict liability or tort arising in any way out of the use of this software, even
The possible training methods (parameter rprop.method) are the
following (see References for details):
| Rprop+: | Rprop with Weight-Backtracking |
| Rprop-: | Rprop without Weight-Backtracking |
| iRprop+: | Improved Rprop with Weight-Backtracking |
| iRprop-: | Improved Rprop without Weight-Backtracking |
M. Riedmiller, H. Braun. A direct adaptive method for faster backpropagation learning: The RPROP algorithm. In Proceedings of the IEEE International Conference on Neural Networks, pp 586-591. IEEE Press, 1993.
C. Igel , M. Huesken. Improving the Rprop Learning Algorithm, Proceedings of the Second International Symposium on Neural Computation, NC 2000, ICSC Academic Press, Canada/Switzerland, pp. 115-121., 2000.
Kohavi, R., A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection, Proceedings of the 14th Int. Joint Conference on Artificial Intelligence 2, S. 1137-1143, Morgan Kaufmann, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1995.
Other fine-tuning functions: backpropagation,
minimizeAutoencoder,
minimizeClassifier
# NOT RUN {
data(iris)
model <- darch(Species ~ ., iris, darch.fineTuneFunction = "rpropagation",
preProc.params = list(method = c("center", "scale")),
darch.unitFunction = c("softplusUnit", "softmaxUnit"),
rprop.method = "iRprop+", rprop.decFact = .5, rprop.incFact = 1.2,
rprop.initDelta = 1/100, rprop.minDelta = 1/1000000, rprop.maxDelta = 50)
# }
Run the code above in your browser using DataCamp Workspace