- The CALIBRATION toolbox
FAST
Fpcrprd:- Prediction using Fast PCR in a PLS-1 basis
Plsf:- Fast PLS-1 calibration procedure
LWR
Lwrcv:- Performs Locally Weighted Regression with leave-one-out CV
to determine (1) the optimal model complexity and (2) the
optimal number of nearest neighbours to be used in future
Lwrev:- Performs Locally Weighted Regression with External Validation
to determine (1) the optimal model complexity and (2) the
optimal number of nearest neighbours to be used in future
Lwrpred:- Predicts Yhat of new samples (Ynew is unknown) using LWR model
with a-components and n_neigh of nearest neighbours determined
by CROSS/EXTERNAL validation (m-files: lwrcv.m, lwrev.m)
MLR
Checkson:- Checks whether the two strings obtained after
cross-over and mutation obey the constraints
of the GA
Checksub: - Checks whether children strings can enter the
population, and replace a string yielding a worse response
Ga0:- Selection of variables for MLR with a genetic algorithm
Ga1:- Main routine of the GA
Gainit:- Random initiation and evaluation of the population
Gamut:- Performs random mutation in each child string
Geomean:- Computation of the geometrical mean of a set
of values, placed in a vector
Mlr:- Computation of the coefficients of a MLR model
Mlrcv:- Cross-validation of a MLR model
Mlrev:- External validation of a MLR model
Mlrpred:- Prediction of a test set with a MLR model
Stepmlr:- Stepwise Multiple Linear Regression Method
Stepson:- Eliminates the 'nb_iter' less informative variables
from a string
Stepwise:- Performs the backward elimination on (one of) the
best string(s) in the actual population (requires SWT01.mat and SWT05.mat)
NNET (requires KDSMIR.mat)
Drawnn:- Draws the topology of a neural network. (Called in nnmodel.m)
Kolmog:- Kolmogorov-Smirnov test to check if a set of observations
are normally distributed
Kolnorm:- Performs a Kolmogorov-Smirnov test for normality on replicate
objects at different levels, and returns the percentage of
objects not normally distributed at each level
Levmarq:- Levenberg-Marquardt algorithm for training feed-forward
backpropagation neural networks . Also pruned (ie. not
fully connected) networks can be trained
Linfit:- Construct a linear regression model to fit a calibration
line and estimate lack-of-fit
Lmeval:- Computes the output of a backpropagation neural network
Nnmodel:- Backpropagation neural network with one hidden layer for
multivariate calibration. (Designed to model only one response y at a time)
Nnpred:- Prediction of responses of new samples with a neural network model
Pmntanh:- Fast hyperbolic tangent function to be used in neural networks
instead of the tanh provided by MATLAB
Rangenew:- Each element in the new data set is scaled to a range determined
by the user on a reference set
PCR
Pcrcv:- Performs cross (internal) validation (CV) of a PCR model
with the aim to determine the optimal model complexity
Pcrev:- Performs external validation of a PCR model to determine
the optimal model complexity
Pcrpred:- Performs prediction of y for new samples using PCR model
built on the inputted calibration X and y data
Pcrsel:- Performs the selection of PCs for PCR modelling
Pcruve:- Performs uninformative variable elimination in the PCR
model with specified order of PCs
Pcruvecv:- VALIDATION of the UVE-PCR model developed on only the
retained variables (pcruvecv.m uses the output variable
"wavsel" of the pcruve.m function)
PLS
Plscv:- PLS cross-validation using the SIMPLS or WIMPLS algorithm,
respectively for tall or wide X-data. The optimal approach is selected automatically.
Performs CROSS-VALIDATION (CV) of PLS model with the
aim to determine the optimal model complexity
Plscvsim:- PLS cross-validation using the SIMPLS algorithm
Plscvwim:- PLS cross-validation using the SIMPLS algorithm
modified for wide X-data. The algorithm uses the small
matrix D=X*X' (n x n) instead of the original wide X
Plsev:- PLS external-validation/evaluation of the predictive ability
Plspert:- Performs INTERNAL VALIDATION of PLS model with the aim to
determine optimal model complexity. (Alternative to CROSS-VALIDATION)
Plspred:- Predicts Yhat of new samples (Ynew is unknown) using PLS model
with a-components, where a was determined by CROSS/EXTERNAL
validation (using the m-file: plscv.m or plsev.m)
Plssim:- Full implementation of SIMPLS approach to PLS regression
for (multivariate) Y
Plsuve:- Simple implementation of the Un-informative Variable Elimination
(UVE)-PLS algorithm
Plsuvecv:- Full implementation of the Un-informative Variable Elimination
(UVE)-PLS algorithm including CROSS-VALIDATION
Rcepls:- Relevant Component Extraction for PLS
STATS (requires KDSMIR.mat)
Durbin:- To examine the serial correlation in a serie of residuals
ranked according to y.
This m-file is a subroutine needed for nonl_det.m
Kolmog:- Kolmogorov-Smirnov test to check if a set of observations
are normally distributed.
Kolnorm:- Performs a Kolmogorov-Smirnov test for normality on replicate
objects at different levels, and returns the percentage of
objects not normally distributed at each level.
Linfit:- Construct a linear regression model to fit a calibration line
and estimate lack-of-fit.
Nonl_det:- detect nonlinearity in the relationship between PC
scores (T) and the property of interest y
Run_test:- To examine the number of runs (series of consecutive
residuals with the same sign) in a set of residuals with
the aim to detect a nonlinear pattern displayed by a
diagnostic (e.g. residual) plot.
This m-file is a subroutine needed for nonl_det.m
