---

AbsoluteNeuralPrediction Class Reference

Neural network based prediction method. More...

#include <prediction.h>

Inheritance diagram for AbsoluteNeuralPrediction:

List of all members.

Public Methods
virtual void	make (const StringMap &params)
virtual void	train (const Matrix &traindata, int startmonth)
virtual Ref< Matrix >	predict (const Matrix &testdata, int startmonth) const
Protected Methods
PatternSet *	makeSet (const Matrix &data, int startmonth) const
Protected Attributes
ANNetwork *	mpNetwork
bool	mUseAllOutputs
bool	mUseAllInputs
int	mVariable
bool	mGlobalEqualization
String	mHiddenTopology
StringMap	mParams
TrainingObserver *	rpObserver

---

Detailed Description

Neural network based prediction method.

Easily customizable by the undocumented parameters for the make() method.

Definition at line 292 of file prediction.h.

---

Member Function Documentation

void make (  const StringMap &    params )  [virtual]

Initialization. Reimplemented from PredictionStrategy. Definition at line 564 of file prediction.cc. References PredictionStrategy::make(), mGlobalEqualization, mHiddenTopology, mParams, mpNetwork, mUseAllInputs, and mUseAllOutputs.

PatternSet * makeSet (  const Matrix &    data, int    startmonth )  const [protected]

Builds pattern set from given dataset and starting month. Definition at line 578 of file prediction.cc. References PredictionStrategy::mInputMonths, mUseAllInputs, mUseAllOutputs, mVariable, PatternSource::patterns, PatternSet::set_input(), and PatternSet::set_output(). Referenced by predict(), and train().

Ref< Matrix > predict (  const Matrix &    testdata, int    startmonth )  const [virtual]

Tests the data and returns the monthly predictions in matrix. Must be implemented by prediction strategies. Parameters: startmonth  See train(). Reimplemented from PredictionStrategy. Definition at line 679 of file prediction.cc. References ANNetwork::getEqualizer(), makeSet(), mpNetwork, mUseAllOutputs, PatternSource::outputs, PatternSource::patterns, and ANNetwork::testPattern().

void train (  const Matrix &    traindata, int    startmonth )  [virtual]

Trains the learning method with the given data. Must be implemented by prediction strategies. Parameters: startmonth  Zero-based month number for the first sample in the data. The month number can be in range 0-11, but only the last two digits are interpreted as months, so it can have values like 199905 or 9905. Reimplemented from PredictionStrategy. Definition at line 621 of file prediction.cc. References MatrixEqualizer::analyze(), ANNetwork::connectFullFfw(), ANNetwork::getEqualizer(), RPropTrainer::init(), ANNetwork::init(), ANNetwork::make(), makeSet(), mGlobalEqualization, mHiddenTopology, mParams, mpNetwork, rpObserver, ANNetwork::setEqualizer(), Trainer::setObserver(), Trainer::setTerminator(), and Trainer::train().

---

Member Data Documentation

bool mGlobalEqualization [protected]

Should global equalization be used? Definition at line 326 of file prediction.h. Referenced by make(), and train().

String mHiddenTopology [protected]

Topology description string for the hidden units, for example "10-5-5". Definition at line 331 of file prediction.h. Referenced by make(), and train().

StringMap mParams [protected]

Application parameters, stored here for subsequent use. Definition at line 334 of file prediction.h. Referenced by make(), and train().

ANNetwork* mpNetwork [protected]

Trained network. Definition at line 307 of file prediction.h. Referenced by make(), predict(), and train().

bool mUseAllInputs [protected]

Should all input variables be used or just one? If false, mUseAllOutputs must also be false (we can't predict all variables with the information from just one variable). Definition at line 318 of file prediction.h. Referenced by make(), and makeSet().

bool mUseAllOutputs [protected]

Flag that says whether we train all variables at a time, i.e., have all the variables in the output layer. Definition at line 312 of file prediction.h. Referenced by make(), makeSet(), and predict().

int mVariable [protected]

Currently handled output variable. This value changes during the run. Definition at line 323 of file prediction.h. Referenced by makeSet().

TrainingObserver* rpObserver [protected]

Training observer that can print out training log during network training. Definition at line 339 of file prediction.h. Referenced by train().

---

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

• prediction.h
• prediction.cc

---