Plot Optimal Model from Classification with Parallel Factor Analysis

Plots optimal model based on results from a 'wrapcpfa' object generated by function cpfa.

Usage

plotcpfa(object, cmeasure = "acc", meanvalue = TRUE, supNum = FALSE, 
         cmode = NULL, parallel = FALSE, cl = NULL, scale.remode = NULL, 
         newscales = 1, scale.abmode = NULL, sign.remode = NULL, newsigns = 1, 
         sign.abmode = NULL, ...)

Arguments

object: An object of class 'wrapcpfa' from function cpfa.
cmeasure: Classification performance measure used to select the optimal number of components. Options include c("err", "acc", "tpr", "fpr", "tnr", "fnr", "ppv", "npv", "fdr", "fom", "fs"). If cmeasure is in c("err", "fpr", "fnr", "fdr", "fom"), the number of components that minimized cmeasure is selected among all classification methods. Otherwise, the number that maximized cmeasure is selected.
meanvalue: Logical indicating whether to find the optimal number of components based on the mean performance across replications from the results generated by cpfa. If meanvalue = FALSE, the median is used.
supNum: Logical indicating whether to suppress text displaying component weight values within plot cells. If TRUE, values are not displayed.
cmode: Integer value of 1, 2, or 3 (or 4 if x is a four-way array) specifying the mode whose component weights were predictors for classification when generating the 'wrapcpfa' object using function cpfa. If the original input model from function cpfa was pca, cmode must be 1 or 2 when the original input x was a two-way matrix. If the original input x was a three-way or four-way array, cmode values greater than 2 are ignored with a warning.
parallel: Logical indicating if parallel computing should be implemented. Defaults to FALSE, which implements sequential computing.
cl: Cluster for parallel computing, which is used when parallel = TRUE. Note that if parallel = TRUE and cl = NULL, then the cluster is defined as makeCluster(max(1L, detectCores() - 1L)).
scale.remode: Character that indicates a mode to rescale. Must be one of c("A", "B", "C", "D"). Sent directly to argument mode in function rescale from package multiway. See help file for rescale for additional details. Ignored when function cpfa argument model = 'pca'. If supplied, argument scale.abmode must also be supplied.
newscales: The root mean-square for columns of the mode indicated by scale.remode. See help file for rescale for additional details. Ignored when function cpfa argument model = 'pca'.
scale.abmode: Character that indicates the mode that absorbs the inverse of rescalings applied to the mode indicated by scale.remode. Must be one of c("A", "B", "C", "D"). Sent directly to argument absorb in function rescale from package multiway. See help file for rescale for additional details. Ignored when function cpfa argument model = 'pca'. If supplied, argument scale.remode must also be supplied.
sign.remode: Character that indicates a mode to resign. Must be one of c("A", "B", "C", "D"). Sent directly to argument mode in function resign from package multiway. See help file for resign for additional details. Ignored when function cpfa argument model = 'pca'. If supplied, argument sign.abmode must also be supplied.
newsigns: Scalar or vector indicating resignings for columns of the mode indicated by sign.remode. See help file for resign for additional details. Ignored when function cpfa argument model = 'pca'.
sign.abmode: Character that indicates the mode that absorbs the negation of the resignings applied to the mode indicated by sign.remode. Must be one of c("A", "B", "C", "D"). Sent directly to argument absorb in function resign from package multiway. See help file for resign for additional details. Ignored when function cpfa argument model = 'pca'. If supplied, argument sign.remode must also be supplied.
...: Additional arguments to be passed to function parafac for fitting a Parafac model or function parafac2 for fitting a Parafac2 model. See help file for function parafac or for function parafac2 for additional details.

Details

Selects the number of components that optimized a performance measure across all classification methods used by cpfa. With this optimal number of components, fits the component model that was used by cpfa to create the input 'wrapcpfa' object. Uses same constraints used in cpfa. Plots component weights for this optimal model using heatmaps. Darker red indicates component weights that are more negative while darker green indicates component weights that are more positive. For three-way Parafac, plots A and B weights. For four-way Parafac, plots A, B, and C weights. For three-way Parafac2, plots B weights. For four-way Parafac2, plots B and C weights. For PCA, plots loadings (i.e., A weights).

Value

Returns one or more heatmap plots of component weights for the optimal component model. Returns list of estimated component weights from the optimal model.

Aweights: Estimated A mode weights for optimal Parafac or Parafac2 model. For PCA, contains the component loadings.
Bweights: Estimated B mode weights for optimal Parafac or Parafac2 model. For PCA, contains the component scores.
Cweights: Estimated C mode weights for optimal Parafac or Parafac2 model. For PCA, returns NULL.
Dweights: Estimated D mode weights for optimal Parafac or Parafac2 model. For PCA, returns NULL.

Note

Applications of this function to real datasets can be explored at the following repository: https://github.com/matthewasisgress/multiway-classification/.

Author

Matthew Asisgress <mattgress@protonmail.ch>

References

See help file for function cpfa for a list of references.

Examples

########## Parafac2 example with 4-way array and multiclass response ##########
if (FALSE) { # \dontrun{
# set seed and simulate a four-way ragged array related to a multiclass response
set.seed(5)

# define list of arguments specifying distributions for A and G weights
techlist <- list(distA = list(dname = "poisson", 
                              lambda = 3),                 # for A weights
                 distG = list(dname = "gamma", shape = 2, 
                              scale = 4))                  # for G weights

# define target correlation matrix for columns of D mode weights matrix
cormat <- matrix(c(1, .6, .6, .6, 1, .6, .6, .6, 1), nrow = 3, ncol = 3)

# simulate a four-way ragged array connected to a response
data <- simcpfa(arraydim = c(10, 11, 12, 100), model = "parafac2", nfac = 3, 
                nclass = 3, nreps = 1e2, onreps = 10, corresp = rep(.6, 3), 
                meanpred = rep(2, 3), modes = 4, corrpred = cormat,
                technical = techlist, smethod = "eigende")

# initialize
alpha <- seq(0, 1, length = 2)
gamma <- c(0, 1)
cost <- c(0.1, 5)
rda.alpha <- seq(0.1, 0.9, length = 2)
delta <- c(0.1, 2)
method <- c("PLR", "SVM", "RDA")
family <- "multinomial"
parameters <- list(alpha = alpha, gamma = gamma, cost = cost, 
                   rda.alpha = rda.alpha, delta = delta)
model <- "parafac2"
nfolds <- 3
nstart <- 3

# constrain first mode weights to be orthogonal, fourth mode to be nonnegative
const <- c("orthog", "uncons", "uncons", "nonneg")

# fit Parafac2 model and use fourth mode weights to tune classification
# methods, to predict class labels, and to return classification 
# performance measures pooled across multiple train-test splits
output <- cpfa(x = data$X, y = data$y, model = model, nfac = 3, 
               nrep = 2, ratio = 0.8, nfolds = nfolds, method = method, 
               family = family, parameters = parameters, 
               type.out = "descriptives", seeds = NULL, plot.out = TRUE, 
               parallel = FALSE, const = const, nstart = nstart, ctol = 1e-2)

# plot heatmap of component weights for optimal model
plotcpfa(output, nstart = nstart, ctol = 1e-3)
} # }