inst/rchemo_functions_github.md

In rchemo: Dimension Reduction, Regression and Discrimination for Chemometrics

R Package rchemo

https://github.com/ChemHouse-group/rchemo

PCA

• pcaeigen Eigen decomposition
• pcaeigenk Eigen for wide matrices (kernel form)
• pcasvd SVD decomposition
• pcanipals Nipals
• pcanipalsna Nipals allowing missing data
• kpca Non Linear Kernel PCA (KPCA) (Scholkopf et al. 2002)
• pcasph Spherical PCA

Auxiliary (works also for PLS)

• scordis Score distances (SD) for a score (T) space
• odis Orthogonal distances (OD) for a score (T) space
• xfit Matrix fitting from score (T) and loadings (P) matrices

REGRESSION

Linear

• lmr Multiple linear regression
• cglsr: CGLSR algorithm for ill-conditionned systems (Bjorck 1996)

Partial least squares

• plskernel "Improved Kernel #1" (Dayal & McGregor 1997)
• plsnipals Nipals
• plsrannar Kernel version for wide matrices (Rannar et al. 1994)
• kplsr Non linear kernel PLSR (KPLSR) (Rosipal & Trejo 2001)
• dkplsr Direct KPLSR (Bennett & Embrechts 2003)

Auxiliary

• dfplsr_cg, dfplsr_cov, dfplsr_div: Model complexity estimation for PLSR
• aicplsr: Cp and AIC for PLSR models
• vip: Variable Importance in Projection (works also for PLSDA models)
• plsr_plsda_allsteps: help to determine the optimal number of latent variables, permutation test, model, predictions for PLSR (works also for PLSDA models)

Ridge

• rr Ridge Regression (RR)
• krr Non linear kernel ridge regression (KRR) = Least-square SVM regression (LS-SVMR)
• dkrr Direct KRR

Support vector machine

• svmr SVM regression (SVMR)

K-nearest-neighbors

• knnr KNN regression (KNNR)
• lwplsr KNN Locally weighted PLSR (KNN-LWPLSR)

Ensemblist methods

Averaging PLSR models with different numbers of LVs

• plsr_agg PLSR-AGG
• lwplsr_agg KNN-LWPLSR-AGG

DISCRIMINATION ANALYSIS

Factorial discrimination analysis (FDA)

• fda Eigen decomposition of the compromise "inter/intra"
• fdasvd Weighted SVD decomposition of the class centers

On predicted Y-dummy table

• lmrda DA on LMR prediction (LMR-DA)
• plsrda DA on PLSR prediction (PLSR-DA = common "PLSDA")
• kplsrda DA on KPLSR prediction (KPLSR-DA)
• rrda DA on RR prediction (RR-DA)
• krrda DA on KRR prediction (KRR-DA)

Probabilistic

• lda Linear discriminant analysis (LDA)
• qda Quadratic discriminant analysis (QDA)
• plslda LDA on PLS latent variables (LVs) (PLS-LDA)
• plsqda QDA on PLS LVs (PLS-QDA)

Support vector machine

• svmda SVMDA (= SVMC)

K-nearest-neighbors

• knnda KNN-DA
• lwplsrda KNN Locally weighted PLSR-DA (KNN-LWPLSR-DA)
• lwplslda KNN Locally weighted PLS-LDA/QDA (KNN-LWPLS-LDA/QDA)

Ensemblist methods

Averaging PLSR models with different numbers of LVs

• plsrda_agg PLSRDA-AGG
• lwplsrda_agg KNN-LWPLSR-DA-AGG
• lwplslda_agg KNN-LWPLS-LDA-AGG
• lwplslda_agg KNN-LWPLS-QDA-AGG

TUNING MODELS

Validation dataset

• gridscore Any model
• gridscorelv Models with LVs (faster)
• gridscorelb Models with ridge parameter (faster)

Cross-validation (CV)

• gridcv Any model
• gridcvlv Models with LVs (faster)
• gridcvlb Models with ridge parameter (faster)

Auxiliary

• segmkf Building segments for K-fold CV
• segmts Building segments for test-set CV

Performance scores

• msep MSEP
• rmsep RMSEP
• sep SEP
• bias Bias
• r2 R2
• cor2 Squared correlation
• rpd, rpdr Ratio of performance to deviation
• mse Summary for regression
• err Classification error rate

Heuristic

• selwold Wold's criterion for models with LVs

SELECTION OF VARIABLES

• covsel COVSEL algorithm (Roger et al. 2011)

DATA MANAGEMENT

Pre-processing

• detrend Polynomial detrend transformation
• snv Standard-normal-deviation transformation
• mavg Smoothing by moving average
• savgol Savitsky-Golay filtering (derivation)
• dderiv Derivation by finite difference
• xinterp Resampling of spectra by interpolation methods
• rmgap Remove vertical gaps in spectra (e.g. for ASD)
• eposvd Pre-processing data by external parameter orthogonalization (EPO; Roger et al 2003)

Sampling

• sampks Kennard-Stone sampling
• sampdp Duplex sampling
• sampcla Within-class stratified sampling

Checking

• checkna Find and count NA values in a data set
• plotxna Plotting missing data in a matrix
• checkdupl Find duplicated row observations between two data sets
• rmdupl Remove duplicated row observations between two data sets

Summary

• aggmean Centers of classes
• dtagg Summary statistics with data subsets
• summ Summary of the quantitative variables of a data set

Multi-block

• mblocks Makes a list of blocks
• hconcat Horizontal block concatenation
• blockscal Block autoscaling
• orthog Orthogonalization of two blocks
• mbplsr, mbplskern, mbplsnipals, mbplsrannar: Multi-Block PLS
• mbplsrda Multi-Block PLS-RDA
• mbplslda Multi-Block PLS-LDA
• mbplsqda Multi-Block PLS-QDA
• sopls Sequential and Orthogonalized PLS
• soplsrda Sequential and Orthogonalized PLS-RDA
• soplslda Sequential and Orthogonalized PLS-LDA
• soplsqda Sequential and Orthogonalized PLS-QDA
• soplscv Help to choose optimal Sequential and Orthogonalized PLS model
• soplsrdacv Help to choose optimal Sequential and Orthogonalized PLS-RDA model
• soplsldacv Help to choose optimal Sequential and Orthogonalized PLS-LDA model
• soplsqdacv Help to choose optimal Sequential and Orthogonalized PLS-QDA model
• mbplsr_mbplsda_allsteps: help to determine the optimal number of latent variables, permutation test, model, predictions for MBPLSR (works also for MBPLSDA models)
• soplsr_soplsda_allsteps: help to determine the optimal number of latent variables, permutation test, model, predictions for SOPLSR (works also for SOPLSDA models)

Datasets

• *asdgap ASD spectra with vertical gaps
• cassav Tropical shrubs
• forages Tropical forages
• octane Gazoline "octane" dataset
• ozone Los Angeles "ozone" pollution (1976) dataset

GRAPHICS

• plotsp Plotting spectra, loadings, or more generally row observations of a data set
• plostsp1 Same as plotsp but one-by-one row
• plotxy 2-d scatter plot
• plotjit Jittered plot
• plotscore Plotting error rates of prediction models

AUXILIARY

• dmnorm Multivariate normal probability density
• dummy Dummy table
• euclsq, *euclsq_mu Euclidean distance matrices
• mahsq, *mahsq_mu Mahalanobis distance matrices
• getknn KNN selection
• krbf, kpol, ktanh Gram matrices for different kernels
• headm Print the first part of a matrix or data frame
• locw Working function for locally weighted models
• matB, matW Between and within covariance matrices
• pinv Moore-Penrose pseudo-inverse
• sourcedir Source every R functions in a directory
• wdist Weights for distances
• Additional working functions in file zfunctions.R

AUTHORS

Marion Brandolini-Bunlon

• Universite Clermont Auvergne, INRAE, UNH, Plateforme d Exploration du Metabolisme, MetaboHUB Clermont, 63000 Clermont-Ferrand, France

• ChemHouse, Montpellier

[e-mail]

Benoit Jaillais

• Oniris, INRAE, StatSC, 44322 Nantes, France

• ChemHouse, Montpellier

[e-mail]

Jean-Michel Roger

• ITAP, INRAE Montpellier Institut Agro, Universite de Montpellier, Montpellier.

• ChemHouse, Montpellier

[e-mail]

Matthieu Lesnoff

• Cirad, UMR Selmet, Montpellier, France

• ChemHouse, Montpellier

[e-mail]

rchemo documentation built on Sept. 11, 2024, 8:05 p.m.