Learn more about Model-Based Calibration

CreateAlgorithm - Create algorithm

Syntax

newalg = alg.CreateAlgorithm( AlgorithmName)

Description

This is a method of mbcmodel.fitalgorithm.

newalg = alg.CreateAlgorithm( AlgorithmName) creates an algorithm of the specified type. alg is a mbcmodel.fitalgorithm object. AlgorithmName must be in the list of alternative algorithms given by alg.getAlternativeNames.

To change the fit algorithm for a model:

>> mdl = mbcmodel.CreateModel('Polynomial', 2);
>> minpress = mdl.FitAlgorithm.CreateAlgorithm('Minimize PRESS');
>> mdl.FitAlgorithm = minpress;

The AlgorithmName determines what properties you can set. You can display the properties for an algorithm as follows:

>> mdl.FitAlgorithm.properties

Algorithm: Minimize PRESS
Alternatives: 'Least Squares','Forward Selection','Backward 
Selection','Prune'
     MaxIter: Maximum Iterations (int: [1,1000])

As a simpler alternative to using CreateAlgorithm, you can assign the algorithm name directly to the algorithm. For example:

B.FitAlgorithm.BoundaryPointOptions = 'Boundary Only';

Or:

m.FitAlgorithm = ‘Minimize PRESS';

Case and spaces are ignored. See FitAlgorithm.

The following sections list the properties available for each algorithm type.

Linear Model Algorithm Properties

Linear Models Algorithms

Used by polynomials, hybrid splines and as the StepAlgorithm for RBF algorithms.

Algorithm: Least Squares

Alternatives: 'Minimize PRESS','Forward Selection','Backward Selection','Prune'

Algorithm: Minimize PRESS

Alternatives: 'Least Squares','Forward Selection','Backward Selection','Prune'

MaxIter: Maximum Iterations (int: [1,1000])

Algorithm: Forward Selection

Alternatives: 'Least Squares','Minimize PRESS','Backward Selection','Prune'

ConfidenceLevel: Confidence level (%) (numeric: [70,100])
MaxIter: Maximum Iterations (int: [1,1000])
RemoveAll: Remove all terms first (Boolean)

Algorithm: Backward Selection

Alternatives: 'Least Squares','Minimize PRESS','Forward Selection','Prune'

ConfidenceLevel: Alpha (%) (numeric: [70,100])
MaxIter: Maximum Iterations (int: [1,1000])
IncludeAll: Include all terms first (Boolean)

Algorithm: Prune

Alternatives: 'Least Squares','Minimize PRESS','Forward Selection','Backward Selection'

Criteria (PRESS RMSE|RMSE|GCV|Weighted PRESS|-2logL|AIC|AICc|BIC|R^2|R^2 adj|PRESS R^2|DW|Cp|cond(J))
MinTerms: Minimum number of terms (int: [0,Inf])
Tolerance (numeric: [0,1000])
IncludeAll: Include all terms before prune (Boolean)
Display (Boolean)

RBF Algorithm Properties

For information about any of the RBF and Hybrid RBF algorithm properties, see Radial Basis Functions, and especially Fitting Routines in the Model Browser User's Guide.

Algorithm: RBF Fit

WidthAlgorithm: Width selection algorithm (mbcmodel.fitalgorithm)
StepAlgorithm: Stepwise (mbcmodel.fitalgorithm)

Width Selection Algorithms

Alternatives: 'WidPerDim','Tree Regression'

Algorithm: TrialWidths

NestedFitAlgorithm: Lambda selection algorithm (mbcmodel.fitalgorithm)
Trials: Number of trial widths in each zoom (int: [2,100])
Zooms: Number of zooms (int: [1,100])
MinWidth: Initial lower bound on width (numeric: [2.22045e-016,1000])
MaxWidth: Initial upper bound on width (numeric: [2.22045e-016,100])
PlotFlag: Display plots (Boolean)
PlotProgress: Display fit progress (Boolean)

Algorithm: WidPerDim

Alternatives: 'TrialWidths','Tree Regression'

NestedFitAlgorithm: Lambda selection algorithm (mbcmodel.fitalgorithm)
DisplayFlag: Display (Boolean)
MaxFunEvals: Maximum number of test widths (int: [1,1e+006])
PlotProgress: Display fit progress (Boolean)

Algorithm: Tree Regression

Alternatives: 'TrialWidths','WidPerDim'

MaxNumRectangles: Maximum number of panels (int: [1,Inf])
MinPerRectangle: Minimum data points per panel (int: [2,Inf])
RectangleSize: Shrink panel to data (Boolean)
AlphaSelectAlg: Alpha selection algorithm (mbcmodel.fitalgorithm)

Lambda Selection Algorithms

Algorithm: IterateRidge

Alternatives: 'IterateRols','StepItRols'

CenterSelectionAlg: Center selection algorithm (mbcmodel.fitalgorithm)
MaxNumIter: Maximum number of updates (int: [1,100])
Tolerance: Minimum change in log10(GCV) (numeric: [2.22045e-016,1])
NumberOfLambdaValues: Number of initial test values for lambda (int: [0,100])
CheapMode: Do not reselect centers for new width (Boolean)
PlotFlag: Display (Boolean)

Algorithm: IterateRols

Alternatives: 'IterateRidge','StepItRols'

CenterSelectionAlg: Center selection algorithm (mbcmodel.fitalgorithm)
MaxNumIter: Maximum number of iterations (int: [1,100])
Tolerance: Minimum change in log10(GCV) (numeric: [2.22045e-016,1])
NumberOfLambdaValues: Number of initial test values for lambda (int: [0,100])
CheapMode: Do not reselect centers for new width (Boolean)
PlotFlag: Display (Boolean)

Algorithm: StepItRols

Alternatives: 'IterateRidge','IterateRols'

MaxCenters: Maximum number of centers (evalstr)
PercentCandidates: Percentage of data to be candidate centers (evalstr)
StartLambdaUpdate: Number of centers to add before updating (int: [1,Inf])
Tolerance: Minimum change in log10(GCV) (numeric: [2.22045e-016,1])
MaxRep: Maximum number of times log10(GCV) change is minimal (int: [1,100])

Center Selection Algorithms

Algorithm: Rols

Alternatives: 'RedErr','WiggleCenters','CenterExchange'

MaxCenters: Maximum number of centers (evalstr)
PercentCandidates: Percentage of data to be candidate centers (evalstr)
Tolerance: Regularized error tolerance (numeric: [2.22045e-016,1])

Algorithm: RedErr

Alternatives: 'Rols','WiggleCenters','CenterExchange'

MaxCenters: Number of centers (evalstr)

Algorithm: WiggleCenters

Alternatives: 'Rols','RedErr','CenterExchange'

MaxCenters: Number of centers (evalstr)
PercentCandidates: Percentage of data to be candidate centers (evalstr)

Algorithm: CenterExchange

Alternatives: 'Rols','RedErr','WiggleCenters'

MaxCenters: Number of centers (evalstr)
NumLoops: Number of augment/reduce cycles (int: [1,Inf])
NumAugment: Number of centers to augment by (int: [1,Inf])

Tree Regression Algorithms

Algorithm: Trial Alpha

Alternatives: 'Specify Alpha'

AlphaLowerBound: Initial lower bound on alpha (numeric: [2.22045e-016,Inf])
AlphaUpperBound: Initial upper bound on alpha (numeric: [2.22045e-016,Inf])
Zooms: Number of zooms (int: [1,Inf])
Trials: Trial alphas per zoom (int: [2,Inf])
Spacing: Spacing (Linear|Logarimthic)
CenterSelectAlg: Center selection algorithm (mbcmodel.fitalgorithm)

Algorithm: Specify Alpha

Alternatives: 'Trial Alpha'

Alpha: Width scale parameter, alpha (numeric: [2.22045e-016,Inf])
NestedFitAlgorithm: Center selection algorithm (mbcmodel.fitalgorithm)

Algorithm: Tree-based Center Selection

Alternatives: 'Generic Center Selection'

ModelSelectionCriteria: Model selection criteria (BIC|GCV)
MaxNumberCenters: Maximum number of centers (evalstr)

Algorithm: Generic Center Selection

Alternatives: 'Tree-based Center Selection'

CenterSelectAlg: Center selection algorithm (mbcmodel.fitalgorithm)

Hybrid RBF Algorithms

Algorithm: RBF Fit

WidthAlgorithm: Width selection algorithm (mbcmodel.fitalgorithm)
StepAlgorithm: Stepwise (mbcmodel.fitalgorithm)

Width Selection Algorithms

Algorithm: TrialWidths

NestedFitAlgorithm: Lambda and term selection algorithm (mbcmodel.fitalgorithm)
Trials: Number of trial widths in each zoom (int: [2,100])
Zooms: Number of zooms (int: [1,100])
MinWidth: Initial lower bound on width (numeric: [2.22045e-016,1000])
MaxWidth: Initial upper bound on width (numeric: [2.22045e-016,100])
PlotFlag: Display plots (Boolean)
PlotProgress: Display fit progress (Boolean)

Nested Fit Algorithms

Algorithm: Twostep

Alternatives: 'Interlace'

MaxCenters: Maximum number of centers (evalstr)
PercentCandidates: Percentage of data to be candidate centers (evalstr)
StartLambdaUpdate: Number of terms to add before updating (int: [1,Inf])
Tolerance: Minimum change in log10(GCV) (numeric: [2.22045e-016,1])
MaxRep: Maximum number of times log10(GCV) change is minimal (int: [1,100])
PlotFlag: Display (Boolean)

Algorithm: Interlace

Alternatives: 'Twostep'

MaxParameters: Maximum number of terms (evalstr)
MaxCenters: Maximum number of centers (evalstr)
PercentCandidates: Percentage of data to be candidate centers (evalstr)
StartLambdaUpdate: Number of terms to add before updating (int: [1,Inf])
Tolerance: Minimum change in log10(GCV) (numeric: [2.22045e-016,1])
MaxRep: Maximum number of times log10(GCV) change is minimal (int: [1,100])

Boundary Model Fit Algorithm Parameters

The following sections list the available fit algorithm parameters for command-line boundary models. The boundary model fit algorithm parameters have the same fit options as the Boundary Editor GUI. For instructions on using these fit options, see Boundary Model Fit Options in the Model Browser documentation.

Ellipsoid

Algorithm: Constraint Fitting

BoundaryPointOptions: Boundary Points (mbcmodel.fitalgorithm)

The boundary points algorithm uses optimization to find the best ellipse. These options are from fmincon.

Algorithm: Boundary Points

Display: Display (none|iter|final)
MaxFunEvals: Maximum function evaluations (int: [1,Inf])
MaxIter: Maximum iterations (int: [1,Inf])
TolFun: Function tolerance (numeric: [1e-012,Inf])
TolX: Variable tolerance (numeric: [1e-012,Inf])
TolCon: Constraint tolerance (numeric: [1e-012,Inf])

Star-shaped

Algorithm: Constraint Fitting

SpecialPointOptions: Special Points (mbcmodel.fitalgorithm)

BoundaryPointOptions: Boundary Points (mbcmodel.fitalgorithm)

ConstraintFitOptions: Constraint Fit (mbcmodel.fitalgorithm)

Star-shaped—Special Points

Algorithm: Star-shaped Points

CenterAlg: Center (mbcmodel.fitalgorithm)

Algorithm alternatives: 'Mean', 'Median', 'Mid Range', 'Min Ellipse', 'User Defined'

For User Defined only: CenterPoint: User-defined center [X1,X2] (vector: NumberOfActiveInputs)

Star-shaped—Boundary Points

You can choose to find boundary points (use Interior) or to assume that all points are on the boundary (use Boundary Only). The interior algorithm then has manual and auto options for the dilation radius and ray casting algorithms.

Algorithm: Boundary Only (no further options)
Algorithm: Interior. Further options:
- DilationRadius (mbcmodel.fitalgorithm)
  - Algorithm: Auto
  - Algorithm: Manual
    - radius: Radius (numeric: [0,Inf])
- RayCasting (mbcmodel.fitalgorithm)
  - Algorithm: From data
  - Algorithm: Manual
    - nrays: Number of Rays (int: [1,Inf])

Star-shaped—Constraint Fit

Algorithm: Star-shaped RBF Fit

Further options:

Transform (None|Log|McCallum)
KernelOpts: RBF Kernel (mbcmodel.fitalgorithm)
Kernel algorithms can be: wendland, multiquadric, recmultiquadric, gaussian, thinplate, logisticrbf. linearrbf, cubicrbf.
You can specify widths and continuity as sub-properties of particular RBF kernels.
- You can set widths for wendland, multiquadric, recmultiquadric, gaussian, logisticrbf. Width: RBF Width (numeric: [1.49012e-008,Inf])
  You can set Continuity for wendland. Cont: RBF Continuity (0|2|4|6)
RbfOpts: RBF Algorithm (mbcmodel.fitalgorithm)
Algorithm: Interpolation. The following are additional settings for interpolating RBF.
- CoincidentStrategy: Coincident Node Strategy (Maximum|Minimum|Mean)
- Algorithm: Algorithm (Direct|GMRES|BICG|CGS|QMR)
- Tolerance: Tolerance (numeric: [0,Inf])
- MaxIt: Maximum number of iterations (int: [1,Inf])

Examples

First get a fitalgorithm object, F, from a model:

M = mbcmodel.CreateModel('Polynomial', 4);
F = M.FitAlgorithm
 
F = 
Algorithm: Least Squares
Alternatives: 'Minimize PRESS','Forward Selection','Backward 
Selection','Prune'
1x1 struct array with no fields.

Then, to create a new algorithm type:

Alg = CreateAlgorithm(F, 'Minimize PRESS')
 
Alg = 
Algorithm: Minimize PRESS
Alternatives: 'Least Squares','Forward Selection','Backward 
Selection','Prune'
    MaxIter: 50

The AlgorithmName determines what properties you can set. You can display the properties for an algorithm as follows:

>> mdl.FitAlgorithm.properties

Algorithm: Minimize PRESS
Alternatives: 'Least Squares','Forward Selection','Backward 
Selection','Prune'
     MaxIter: Maximum Iterations (int: [1,1000])

As a simpler alternative to using CreateAlgorithm, you can assign the algorithm name directly to the algorithm. For example:

B.FitAlgorithm.BoundaryPointOptions = 'Boundary Only';

Or:

m.FitAlgorithm = ‘Minimize PRESS';

Case and spaces are ignored.

CreateAlgorithm - Create algorithm

Syntax

Description

Linear Model Algorithm Properties

Linear Models Algorithms

RBF Algorithm Properties

Width Selection Algorithms

Lambda Selection Algorithms

Center Selection Algorithms

Tree Regression Algorithms

Hybrid RBF Algorithms

Width Selection Algorithms

Nested Fit Algorithms

Boundary Model Fit Algorithm Parameters

Ellipsoid

Star-shaped

Star-shaped—Special Points

Star-shaped—Boundary Points

Star-shaped—Constraint Fit

Examples

See Also

Recommended Products