PLSRBin function

Partial Least Squares Regression with several Binary Responses

Fits Partial Least Squares Regression with several Binary Responses

PLSRBin(Y, X, S = 2, InitTransform = 5, grouping = NULL, 
tolerance = 5e-05, maxiter = 100, show = FALSE, penalization = 0.1, 
cte = TRUE, OptimMethod = "CG", Multiple = FALSE)

Arguments

• Y: The response
• X: The matrix of independent variables
• S: The Dimension of the solution
• InitTransform: Initial transform for the X matrix
• grouping: Grouping variable when the inial transformation is standardization within groups.
• tolerance: Tolerance for convergence of the algorithm
• maxiter: Maximum Number of iterations
• show: Show the steps of the algorithm
• penalization: Penalization for the Ridge Logistic Regression
• cte: Should a constant be included in the model?
• OptimMethod: Optimization methods from optim
• Multiple: The responses are the indicators of a multinomial variable?

Details

The function fits the PLSR method for the case when there is a set binary dependent variables, using logistic rather than linear fits to take into account the nature of responses. We term the method PLS-BLR (Partial Least Squares Binary Logistic Regression). This can be considered as a generalization of the NIPALS algorithm when the responses are all binary.

Returns

• Method: Description of 'comp1'

• X: The predictors matrix

• Y: The responses matrix

• Initial_Transformation: Initial Transformation of the X matrix

• ScaledX: The scaled X matrix

• tolerance: Tolerance used in the algorithm

• maxiter: Maximum number of iterations used

• penalization: Ridge penalization

• IncludeConst: Is the constant included in the model?

• XScores: Scores of the X matrix, used later for the biplot

• XLoadings: Loadings of the X matrix

• YScores: Scores of the Y matrix

• YLoadings: Loadings of the Y matrix

• Coefficients: Regression coefficients

• XStructure: Correlations among the X variables and the PLS scores

• Intercepts: Intercepts for the Y loadings

• LinTerm: Linear terms for each response

• Expected: Expected probabilities for the responses

• Predictions: Binary predictions of the responses

• PercentCorrect: Global percent of correct predictions

• PercentCorrectCols: Percent of correct predictions for each column

• Maxima: Column with the maximum probability. Useful when the responses are the indicators of a multinomial variable

References

Ugarte Fajardo, J., Bayona Andrade, O., Criollo Bonilla, R., Cevallos‐Cevallos, J., Mariduena‐Zavala, M., Ochoa Donoso, D., & Vicente Villardon, J. L. (2020). Early detection of black Sigatoka in banana leaves using hyperspectral images. Applications in plant sciences, 8(8), e11383.

Author(s)

José Luis Vicente Villardon

Examples

X=as.matrix(wine[,4:21])
Y=cbind(Factor2Binary(wine[,1])[,1], Factor2Binary(wine[,2])[,1])
rownames(Y)=wine[,3]
colnames(Y)=c("Year", "Origin")
pls=PLSRBin(Y,X, penalization=0.1, show=TRUE, S=2)