RVFL Jackknife+ Regressor

Random Vector Functional Link (RVFL) regressor with closed-form Jackknife+ prediction intervals.

RVFLJackknifePlus(
  n_hidden = 200L,
  lambda = 1,
  activation = c("tanh", "relu", "sigmoid"),
  random_state = 0L,
  symmetric = FALSE,
  venv_path = "./venv",
  ...
)

Arguments

n_hidden

int. Number of random hidden features. Set to 0 for ridge regression.

lambda

numeric. Ridge regularization parameter.

activation

character. Activation function for the random hidden layer. One of "tanh", "relu", or "sigmoid".

random_state

integer. Random seed used to generate the hidden layer.

symmetric

logical. If TRUE, symmetric Jackknife+ intervals are returned. Otherwise, asymmetric Jackknife+ intervals are used.

venv_path

character. Path to the Python virtual environment.

Value

An object of class RVFLJackknifePlus.

Examples

if (FALSE) { # \dontrun{
library(datasets)

X <- as.matrix(mtcars[, -1])
y <- mtcars[, 1]

n <- nrow(X)
set.seed(123)
train_index <- sample(seq_len(n), floor(0.8 * n))

X_train <- X[train_index, ]
y_train <- y[train_index]

X_test <- X[-train_index, ]

obj <- mlsauce::RVFLJackknifePlus()

print(obj$get_params())

obj$fit(X_train, y_train)

pred <- obj$predict(X_test)

pred_pi <- obj$predict(
  X_test,
  alpha = 0.05,
  return_pi = TRUE
)

print(pred)
print(pred_pi$lower)
print(pred_pi$upper)
} # }