#include <rprop.h>
Inheritance diagram for RPropTrainer:
Public Methods | |
virtual Array< DynParameter > * | parameters () const |
virtual void | init (const StringMap ¶ms) |
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 |
Design Patterns: Template Method (various parts of the algorithm can be overloaded).
Definition at line 44 of file rprop.h.
|
Implementation for BackpropTrainer.
Reimplemented from BackpropTrainer. Definition at line 88 of file rprop.cc. References BackpropTrainer::backpropagate(), BackpropTrainer::mError, and mGradient. |
|
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 *
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(). |
|
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. |
|
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(). |
|
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(). |
|
Initial per-weight delta.
Definition at line 62 of file rprop.h. Referenced by init(), and initTrain(). |
|
Maximum per-weight delta.
Definition at line 65 of file rprop.h. Referenced by init(), and updateWeights(). |
|
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(). |
|
Per-weight ... We store these here, because we don't want to alter the network objects just because of the training algorithm. |