A caret wrapper for the honeycombHIVE function,
enabling seamless integration with the caret package for
hyperparameter tuning, cross-validation, and performance evaluation.
Value
A caret model definition list. Pass it to
train for model training and evaluation.
Parameters
nAntibodies: Number of initial antibodies in the network.beta: Clone multiplier controlling the number of clones per antibody.epsilon: Threshold for network suppression to remove redundant antibodies.layers: Number of hierarchical layers for iterative refinement.refineOptimizer: Optimizer for gradient-based refinement (e.g. "sgd", "momentum", "adagrad", "adam", "rmsprop").refineSteps: Number of gradient update steps in refinement.refineLR: Learning rate for refinement.refineHuberDelta: Delta parameter used if the "huber" loss is chosen.
Supported Tasks
"Regression": Predicts numeric target values."Classification": Assigns class labels to input observations."Clustering": Groups data points based on similarity (though typicallycaretis used for supervised tasks).
Examples
if (FALSE) { # \dontrun{
library(caret)
# Example: Classification with Iris
data(iris)
X <- as.matrix(iris[, 1:4])
y <- iris$Species
train_control <- trainControl(method = "cv", number = 5)
set.seed(42)
model <- train(
x = X,
y = y,
method = honeycombHIVEmodel,
trControl = train_control,
tuneGrid = expand.grid(
nAntibodies = c(10, 20),
beta = c(3, 5),
epsilon = c(0.01, 0.05),
layers = c(1, 2),
refineOptimizer = "adam",
refineSteps = 5,
refineLR = 0.01,
refineHuberDelta = 1.0
)
)
print(model)
} # }