Properties by Category
Local Model Toolbox
Contents
Data Set Properties
- input - (N x p) Matrix of the inputs
- scaleInput - (class) Scale input object
- output - (N x q) Matrix of the outputs.
- scaleOutput - (class) Scale output object. Empty for no
- dataWeighting - Weighting of the inputs.
- outputWeighting - Weighting of the outputs.
- validationInput - (? x p) Matrix of the validation data inputs.
- validationOutput - (? x q) Matrix of the validation data outputs.
- validationOutput - (? x p) Matrix of the test data inputs.
- testOutput - (? x q) Matrix of the test data outputs.
- unscaledInput - (N x p) Matrix of the unscaled inputs
- unscaledOutput - (N x q) Matrix of the unscaled outputs
Termination Criterion Properties
- maxNumberOfLM - (1 x 1) Maximum number of LLMs termination criterion
- minError - (1 x 1) Loss function threshold termination criterion
- maxNumberOfParameters - (1 x 1) Maximum number of Parameter termination criterion
- maxTrainTime - (1 x 1) Maximum allowed time for the calculation
- maxIterations - (1 x 1) Maximum number of iterations termination criterion
- maxPenaltyDeterioration - (1 x 1) Maximum number of deterioration of the penalty loss function
Lossfunction Properties
- lossFunctionGlobal - Global lossfunction type ('MSE','RMSE','NMSE','NRMSE', 'MISCLASS' and 'R2' can be chosen).
- lossFunctionLocal - Local lossfunction type ('SE','RSE','DRSE' and 'MISCLASS' can be chosen).
- noiseVariance - This property allows the user to define a certain process noise variance, based on prior knowledge about the process noise.
- complexityPenalty - Complexity penalty.
- estNoiseVariance -
- crossvalidationValues - Crossvalidation values for different model complexities
Other Properties
Global Model Properties
- estimationProcedure - Precedure that is used for estimation
- xRegressorDegree - order of the polynomial within the regressor
- xRegressorMaxPower - highest power within the regressor for polynomials
- xRegressorExponentMatrix - exponent matrix for the regressors
- xRegressorType - Type of regressor; 'polynomial' (default) or 'sparsePolynomial'
- xInputDelay - Delays of the input regressors for rule consequents
- xOutputDelay - Delays of the output regressors for rule consequents
- zInputDelay - Delays of the input regressors for rule premises
- zOutputDelay - Delays of the input regressors for rule premises
- smoothness - Global smoothness value for all local model transitions
- kStepPrediction - (scalar) 0 or 1 for static model or one-step-predictor, inf for dynamic calculation
- leafModels - (1 x M) logical vector, true for active models, false for inactive models
- localModels - (1 x M) cell array of all local models objects
- numberOfInputs - (1 x 1) number of all physical inputs, set in training
- numberOfOutputs - (1 x 1) number of all physical outputs, set in training
Local Model Properties
LOLIMOT
- center - (G x nz) center of the gaussian
- lowerLeftCorner - (G x nz) lower left corner of the rectangle surrounding the gaussian
- localLossFunctionValue - (1 x 1) local loss function value
- standardDeviation - (G x nz) standard deviations of all gaussians
- upperRightCorner - (G x nz) upper right corner of the rectangle surrounding the gaussian
- parameter - (nx x 1) parameter of the local model
- zLowerBound - (1 x nz) lower boundary of the training data
- zUpperBound - (1 x nz) upper boundary of the training data
HILOMOT
- center - (1 x nz) vector of the center coordinates
- parameter - (nx x q) vector/matrix of local model parameters
- localLossFunctionValue - (scalar) the local loss function value
- parent - (scalar) index of parent knot
- children - (1 x ?) indices of children knots
- splittingParameter - (nz+1 x 1) parameter of the sigmoid
- localSmoothness - (scalar) smoothness parameter
- pseudoInv -
Data Set Information Properties
History Properties
- currentNumberOfParameters - (1 x iter) Overall number of model parameters for each iteration
- currentNumberOfEffParameters - (1 x iter) Effective number of model parameters for each iteration
- currentNumberOfLMs - (1 x iter) Overall number of models for each iteration
- globalLossFunction - (1 x iter) Loss function values for the overall model
- penaltyLossFunction - (1 x iter) Complexity penalized loss function values
- CVLossFunction -
- splitLM - (iter-1 x 1) History of LMs which are split
- splitDimension - (iter-1 x 1) History of dimensions of the split
- splitRatio - (iter-1 x 1) History of ratio of the split
- leafModelIter - (1 x iter) History of the active models per iteration
- trainingTime - (1 x iter) Overall Training time up to this iteration
- displayMode - (logical) Flag to de-/activate printouts to the command window during training
- iteration - (1 x 1) Stores the number of the used iteration
- validationDataLossFunction - (1 x iter) Validation data loss function values for the overall model
- testDataLossFunction - (1 x iter) Test data loss function values for the overall model
Algorithm Specific Properties
LOLIMOT
- numberOfLMReliable - (1 x 1) Number of LMs for a reliable model estimation
- suggestedNet - (1 x 1) Suggested net with best performance/complexity trade-off
- idxAllowedLM - (1 x M) Stores the allowed local models with respect to splitting
- splits - (1 x 1) Number of splits tested in each dimension
- outputModel - (N x q) Model outputs
- xRegressor - (N x nx) Regression matrix/Regressors for rule consequents
- zRegressor - (N x nz) Regression matrix/Regressors for rule premises
- MSFValue - (1 x M) Membership function values for each local model
HILOMOT
- oblique - logical 0: axes-orthogonal partitioning, 1: axis-oblique partitioning
- numberOfPoints - (1 x 1) Minimum number of points that have to be in a LM before it is allowed for splitting
- optGrad - logical 0: numerical gradient, 1: analytical gradient
- optLOOCV - logical If true, split opt. w.r.t. LOO cross val. error
- suggestedNet - (1 x 1) Suggested net with best performance/complexity trade-off
- inputSensitivity - (nz x 1) Sensitivity analysis of zRegressor-inputs. Value 1 has highest importance, value 0 seems to be a redundant input.
- outputModel - (N x q) Model outputs
- xRegressor - (N x nx) Regression matrix/Regressors for rule consequents
- zRegressor - (N x nz) Regression matrix/Regressors for rule premises
- phi - (N x M) Validity function matrix
SYMBOLS AND ABBREVIATIONS
LM: Local model
p: Number of inputs (physical inputs)
q: Number of outputs
N: Number of data samples
M: Number of LMs
nx: Number of regressors (x)
nz: Number of regressors (z)
% * <_help.html ||> -