You are here: Start » FIL.NET » Function Reference » Data Classification » Multilayer Perceptron » FIL.MLP_Train

FIL.MLP_Train

Creates and trains multilayer perceptron classifier.

Namespace:FilNet
Assembly:FIL.NET.dll

Syntax

C++
C#
 
public static void MLP_Train
(
	FilNet.MlpModel inMlpModel,
	IList<IList<float>> inInputVectorArray,
	IList<IList<float>> inResponseVectorArray,
	NullableRef<IList<IList<float>>> inTestInputVectorArray,
	NullableRef<IList<IList<float>>> inTestResponseVectorArray,
	int inIterationCount,
	float inLearningRate,
	float inMomentum,
	int? inRandomSeed,
	FilNet.MlpModel outMlpModel,
	FilNet.Profile diagErrorChartLearning,
	FilNet.Profile diagErrorChartTesting
)

Parameters

Name Type Range Default Description
inMlpModelFilNet.MlpModelInitialized MLP model.
inInputVectorArraySystem.Collections.Generic.IList<System.Collections.Generic.IList<float>>Array of features used to train model.
inResponseVectorArraySystem.Collections.Generic.IList<System.Collections.Generic.IList<float>>Array of answers which classifier should get.
inTestInputVectorArrayFilNet.NullableRef<System.Collections.Generic.IList<System.Collections.Generic.IList<float>>>Array of features used to test classifier during training process. Default value: ftl::NIL.
inTestResponseVectorArrayFilNet.NullableRef<System.Collections.Generic.IList<System.Collections.Generic.IList<float>>>Array of answers used to test classifier during training process. Default value: ftl::NIL.
inIterationCountint<1, INF>100Learning iteration count. Default value: 100.
inLearningRatefloat<0.01f, 1.0f>1.0fLearning factor. Default value: 1.0f.
inMomentumfloat<0.0f, 1.0f>0.01fLearning momentum ratio. Default value: 0.01f.
inRandomSeedint?<0, INF>Number used as starting random seed. Default value: ftl::NIL.
outMlpModelFilNet.MlpModelTrained MlpModel.
diagErrorChartLearningFilNet.ProfileMean error of testing results data during learning process.
diagErrorChartTestingFilNet.ProfileMean error during learning process.

Description

The filter trains multilayer perceptron classifier. The inInputVectorArray contains an array of data used to train classifier.

The size of input vector should be constant for each provided input array. Each input vector size must be the same as input count provided in MLP_Init filter during classifier initialization.

The inResponseVectorArray contains answer for each data vector provided in inInputVectorArray. Size of all response vectors should be the same and equal to output count set in MLP_Init.

The inLearningRate determines step size during following function gradient. Too big step size may cause miss of optimization function minimum. Small values may cause learning process too long.

The parameter inMomentum defines how learning step results should depend on previous step results.

The inIterationCount specifies the number of iterations of the learning process.

Errors

List of possible exceptions:

Error type Description
DomainError Different sizes of answer vectors in inResponseVectorArray in MLP_Train.
DomainError Different sizes of feature vector array and answer vector array on input in MLP_Train.
DomainError Different sizes of feature vectors in inInputVectorArray in MLP_Train.
DomainError Empty feature array on input in MLP_Train.
DomainError Empty inInputVectorArray in MLP_Train
DomainError Empty inResponseVectorArray in MLP_Train
DomainError Incorrect or uninitialized MlpModel in MLP_Train.
DomainError Using uninitialized MlpModel in MLP_Train
DomainError Wrong size of answer vector in inTestResponseVectorArray in MLP_Train.
DomainError Wrong size of feature vector in inTestInputVectorArray in MLP_Train.

Function Overrides

See also