---

RPropTrainer Class Reference

Resilient error backpropagation algorithm by Riedmiller. More...

#include <rprop.h>

Inheritance diagram for RPropTrainer:

List of all members.

Public Methods
virtual Array< DynParameter > *	parameters () const
virtual void	init (const StringMap &params)
Protected Methods
virtual void	initTrain (ANNetwork &network) const
virtual void	backpropagate (ANNetwork &network, const PatternSet &set, int p) const
virtual void	updateWeights (ANNetwork &network) const
Protected Attributes
double	mDelta0
double	mDeltaMax
Vector	mDelta
Vector	mOldDeltaW
Vector	mGradient

---

Detailed Description

Resilient error backpropagation algorithm by Riedmiller.

Design Patterns: Template Method (various parts of the algorithm can be overloaded).

Definition at line 44 of file rprop.h.

---

Member Function Documentation

void backpropagate (  ANNetwork &    network, const PatternSet &    set, int    p )  const [protected, virtual]

Implementation for BackpropTrainer. Reimplemented from BackpropTrainer. Definition at line 88 of file rprop.cc. References BackpropTrainer::backpropagate(), BackpropTrainer::mError, and mGradient.

void init (  const StringMap &    params )  [virtual]

This file is part of the Inanna library. * * Copyright (C) 1997-2002 Marko Grönroos <magi@iki.fi> * * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Library General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Library General Public License for more details. * * You should have received a copy of the GNU Library General Public * License along with this library; see the file COPYING.LIB. If * not, write to the Free Software Foundation, Inc., 59 Temple Place * Suite 330, Boston, MA 02111-1307, USA. * * Reimplemented from BackpropTrainer. Definition at line 38 of file rprop.cc. References Trainer::init(), BackpropTrainer::mBatchLearning, BackpropTrainer::mDecay, mDelta0, and mDeltaMax. Referenced by AbsoluteNeuralPrediction::train().

void initTrain (  ANNetwork &    network )  const [protected, virtual]

Implementation for BackpropTrainer. Initializes training. Reimplemented from BackpropTrainer. Definition at line 62 of file rprop.cc. References BackpropTrainer::initTrain(), mDelta, mDelta0, mGradient, and BackpropTrainer::mWeightDeltas.

void updateWeights (  ANNetwork &    network )  const [protected, virtual]

Updates weights after backpropagation phase. Reimplemented from BackpropTrainer. Definition at line 105 of file rprop.cc. References BackpropTrainer::mDecay, mDelta, mDeltaMax, mGradient, BackpropTrainer::mWeightDeltas, Connection::setWeight(), and Connection::weight().

---

Member Data Documentation

Vector mDelta [protected]

Per-weight deltas. We store these here, because we don't want to alter the network objects just because of the training algorithm. Definition at line 72 of file rprop.h. Referenced by initTrain(), and updateWeights().

double mDelta0 [protected]

Initial per-weight delta. Definition at line 62 of file rprop.h. Referenced by init(), and initTrain().

double mDeltaMax [protected]

Maximum per-weight delta. Definition at line 65 of file rprop.h. Referenced by init(), and updateWeights().

Vector mGradient [protected]

Per-weight summed error. We store these here, because we don't want to alter the network objects just because of the training algorithm. Definition at line 86 of file rprop.h. Referenced by backpropagate(), initTrain(), and updateWeights().

Vector mOldDeltaW [protected]

Per-weight ... We store these here, because we don't want to alter the network objects just because of the training algorithm. Definition at line 79 of file rprop.h.

---

The documentation for this class was generated from the following files:

• rprop.h
• rprop.cc

---