Click or drag to resize

NMFGdClsUpdate Class

Class NMFGdClsUpdate encapsulates the Gradient Descent - Constrained Least Squares (GDCLS) algorithm for Nonnegative Matrix Facotorization (NMF).
Inheritance Hierarchy
SystemObject
  CenterSpace.NMath.CoreNMFGdClsUpdate

Namespace: CenterSpace.NMath.Core
Assembly: NMath (in NMath.dll) Version: 7.4
Syntax
[SerializableAttribute]
public class NMFGdClsUpdate : INMFUpdateAlgorithm, 
	ICloneable

The NMFGdClsUpdate type exposes the following members.

Constructors
 NameDescription
Public methodNMFGdClsUpdate Creates the update with default lambda of zero.
Public methodNMFGdClsUpdate(Double) Creates the update with specified lambda value.
Top
Properties
 NameDescription
Public propertyLambda Returns the lambda value.
Top
Methods
 NameDescription
Public methodClone Creates a deep copy of this NMFGdClsUpdate.
Public methodCost Returns the value of the cost function for the non-negative factorization of V = WH. The cost function is minimized by the factorization.
Public methodInitialFactors Provides the initial factors for the iteration process that computes W and H in the non-negative matrix factorization V = WH.
Public methodIterationStep Updates the factors W and H in the non-negative matrix factorization V = WH.
Top
Fields
 NameDescription
Protected fieldStatic memberDEFAULT_LAMBDA Default lambda.
Protected fieldlambda_ Lambda.
Top
Remarks
In some cases it may be desirable to enforce a statistical sparsity constraint on the H matrix. As the sparsity of H increases, the basis vectors become more localized, i.e, the parts-based representation of the data in W becomes more and more enhanced. The Gradient Descent - Constrained Least Squares (GDCLS) algorithm enforces sparsity in H using a scheme that penalizes the number of non-zero entries in H. It is a hybrid algorithm that uses the multiplicative update rule for updating W while H is calculated using a constrained least squares model as the metric.
See Also