Package 'weightederm' reference manual

Title:	Weighted Empirical Risk Minimization for Changepoint Regression
Description:	R interface to the 'weightederm' package for 'Python', which provides 'scikit-learn'-style estimators for offline change point regression (data segmentation) via weighted empirical risk minimization. Supports least-squares, Huber, and logistic losses with fixed or cross-validated numbers of change points. Wraps 'Python' via 'reticulate'. Arpino and Venkataramanan (2026) <doi:10.48550/arXiv.2604.11746>.
Authors:	Gabriel Arpino [aut, cre]
Maintainer:	Gabriel Arpino <[email protected]>
License:	Apache License (>= 2.0)
Version:	0.1.0
Built:	2026-05-28 14:54:41 UTC
Source:	https://github.com/gabrielarpino/weightederm-r

Extract last-segment coefficients

Description

Returns the coefficient vector of the unweighted base-loss refit on the last detected segment (the same coefficients used by predict()).

Usage

## S3 method for class 'werm_fit'
coef(object, ...)
## S3 method for class 'werm_fit'
coef(object, ...)

Arguments

object

A werm_fit object.

...

Ignored.

Value

Named numeric vector of length p.

Predict using the last-segment refit

Description

Evaluates the unweighted base-loss model fitted on the last detected segment. For logistic estimators, returns class labels.

Usage

## S3 method for class 'werm_fit'
predict(object, newdata, type = "class", ...)
## S3 method for class 'werm_fit'
predict(object, newdata, type = "class", ...)

Arguments

object

A werm_fit object returned by any ⁠werm_*⁠ function.

newdata

Numeric matrix of shape ⁠(m, p)⁠.

type

Character. For logistic estimators: "class" (default) or "prob" (returns a ⁠(m, 2)⁠ probability matrix).

...

Ignored.

Value

Numeric vector of length m (regression) or character vector / probability matrix (logistic).

Summarise a fitted WERM model

Description

Summarise a fitted WERM model

Usage

## S3 method for class 'werm_fit'
summary(object, ...)
## S3 method for class 'werm_fit'
summary(object, ...)

Arguments

object

A werm_fit object.

...

Ignored.

Value

Invisibly returns object. Called for its printed side-effect.

Configure which Python environment weightederm uses

Description

A thin wrapper around reticulate::use_python() that must be called before any ⁠werm_*⁠ function if the default Python does not have weightederm installed.

Usage

weightederm_configure_python(python, required = TRUE)
weightederm_configure_python(python, required = TRUE)

Arguments

python

Path to the Python binary or virtual-environment directory.

required

Passed to reticulate::use_python().

Value

Invisibly returns NULL. Called for side effects only.

Examples

if (nzchar(Sys.which("python"))) {
  weightederm_configure_python(Sys.which("python"), required = FALSE)
}

if (nzchar(Sys.which("python"))) {
  weightederm_configure_python(Sys.which("python"), required = FALSE)
}

Fit a WERM changepoint model with Huber loss

Description

Like werm_least_squares() but uses the Huber loss, which is more robust to outliers than squared loss.

Usage

werm_huber(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  epsilon = 1.35,
  max_iter = 100L,
  tol = 1e-05,
  penalty = "none",
  alpha = 0
)
werm_huber(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  epsilon = 1.35,
  max_iter = 100L,
  tol = 1e-05,
  penalty = "none",
  alpha = 0
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Numeric vector of length n. Ordered responses.

num_chgpts

Integer. Number of changepoints to detect.

delta

Integer. Minimum gap between candidate changepoints during search. Default 1. Rule of thumb: max(1L, as.integer(0.05 * nrow(X))).

search_method

Character. "efficient" (default) or "brute_force".

fit_intercept

Logical. Whether each segment model includes an intercept. Default TRUE.

epsilon

Numeric. Huber transition parameter. Default 1.35 (95 % Gaussian efficiency).

max_iter

Integer. Maximum L-BFGS-B iterations. Default 100.

tol

Numeric. Gradient-norm tolerance. Default 1e-5.

penalty

Character. "none" (default), "l1", or "l2".

alpha

Numeric. Penalty strength. Default 0.

Value

An object of class c("werm_huber", "werm_fit"). Same elements as werm_least_squares().

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMHuber")) {
  set.seed(2)
  n <- 12L; p <- 1L; true_cp <- 6L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:true_cp, , drop = FALSE] %*% 3,
    X[(true_cp + 1L):n, , drop = FALSE] %*% -3
  ) + rnorm(n, sd = 0.03)
  fit <- werm_huber(X, y, num_chgpts = 1L, delta = 1L,
                    fit_intercept = FALSE, max_iter = 5L, tol = 1e-3)
  fit$changepoints
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMHuber")) {
  set.seed(2)
  n <- 12L; p <- 1L; true_cp <- 6L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:true_cp, , drop = FALSE] %*% 3,
    X[(true_cp + 1L):n, , drop = FALSE] %*% -3
  ) + rnorm(n, sd = 0.03)
  fit <- werm_huber(X, y, num_chgpts = 1L, delta = 1L,
                    fit_intercept = FALSE, max_iter = 5L, tol = 1e-3)
  fit$changepoints
}

Fit a WERM changepoint model with Huber loss and CV selection

Description

Fit a WERM changepoint model with Huber loss and CV selection

Usage

werm_huber_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  epsilon = 1.35,
  max_iter = 100L,
  tol = 1e-05,
  use_base_loss_for_cv = FALSE,
  penalty = "none",
  alpha = 0
)
werm_huber_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  epsilon = 1.35,
  max_iter = 100L,
  tol = 1e-05,
  use_base_loss_for_cv = FALSE,
  penalty = "none",
  alpha = 0
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Numeric vector of length n. Ordered responses.

max_num_chgpts

Integer. Upper bound of the CV search grid.

delta

Integer. Minimum gap between candidate changepoints. Default 1.

search_method

Character. "efficient" (default) or "brute_force".

cv

Integer. Number of interleaved folds. Default 5.

fit_intercept

Logical. Default TRUE.

epsilon

Numeric. Huber transition parameter. Default 1.35.

max_iter

Integer. Maximum L-BFGS-B iterations. Default 100.

tol

Numeric. Gradient-norm tolerance. Default 1e-5.

use_base_loss_for_cv

Logical. If TRUE, uses squared loss for CV segment fits and scoring instead of the default absolute error. Default FALSE.

penalty

Character. Passed to the inner fixed model. Default "none".

alpha

Numeric. Penalty strength. Default 0.

Value

An object of class c("werm_huber_cv", "werm_fit").

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMHuberCV")) {
  set.seed(11)
  n <- 24L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:8, ] %*% c(4, 0),
    X[9:16, ] %*% c(-4, 0),
    X[17:24, ] %*% c(4, 0)
  ) + rnorm(n, sd = 0.02)
  fit <- werm_huber_cv(X, y, max_num_chgpts = 2L, cv = 3L, delta = 2L,
                       fit_intercept = FALSE, max_iter = 20L)
  fit$best_num_chgpts
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMHuberCV")) {
  set.seed(11)
  n <- 24L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:8, ] %*% c(4, 0),
    X[9:16, ] %*% c(-4, 0),
    X[17:24, ] %*% c(4, 0)
  ) + rnorm(n, sd = 0.02)
  fit <- werm_huber_cv(X, y, max_num_chgpts = 2L, cv = 3L, delta = 2L,
                       fit_intercept = FALSE, max_iter = 20L)
  fit$best_num_chgpts
}

Fit a WERM changepoint model with squared loss

Description

Detects num_chgpts changepoints in ordered regression data by minimising a Weighted Empirical Risk with squared loss.

Usage

werm_least_squares(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  fit_solver = "direct",
  penalty = "none",
  alpha = 0
)
werm_least_squares(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  fit_solver = "direct",
  penalty = "none",
  alpha = 0
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Numeric vector of length n. Ordered responses.

num_chgpts

Integer. Number of changepoints to detect.

delta

Integer. Minimum gap between candidate changepoints during search. Default 1. Rule of thumb: max(1L, as.integer(0.05 * nrow(X))).

search_method

Character. "efficient" (default) or "brute_force".

fit_intercept

Logical. Whether each segment model includes an intercept. Default TRUE.

fit_solver

Character. "direct" (default, closed-form) or "lbfgsb" (gradient-based).

penalty

Character. "none" (default), "l1", or "l2".

alpha

Numeric. Penalty strength. Default 0.

Value

An object of class c("werm_least_squares", "werm_fit") with the following named elements:

changepoints: Integer vector of detected changepoints (1-indexed, R convention). Length equals num_chgpts.
num_chgpts: Integer. Number of detected changepoints.
num_signals: Integer. Number of segments (num_chgpts + 1).
objective: Numeric. Minimised WERM objective value.
last_segment_coef: Numeric vector. Coefficient of the unweighted base-loss refit on the last segment (used by predict()).
last_segment_intercept: Numeric or NULL.
signal_coefs: Matrix ⁠(num_signals x p)⁠. Stage-1 WERM coefficient estimates.
signal_intercepts: Numeric vector or NULL.
n_features_in: Integer. Number of features.

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLeastSquares")) {
  set.seed(1)
  n <- 24L; p <- 2L; true_cp <- 12L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:true_cp, ] %*% c(3, -1.5),
    X[(true_cp + 1L):n, ] %*% c(-3, 1.5)
  ) + rnorm(n, sd = 0.05)
  fit <- werm_least_squares(X, y, num_chgpts = 1L, delta = 3L,
                            fit_intercept = FALSE)
  fit$changepoints
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLeastSquares")) {
  set.seed(1)
  n <- 24L; p <- 2L; true_cp <- 12L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:true_cp, ] %*% c(3, -1.5),
    X[(true_cp + 1L):n, ] %*% c(-3, 1.5)
  ) + rnorm(n, sd = 0.05)
  fit <- werm_least_squares(X, y, num_chgpts = 1L, delta = 3L,
                            fit_intercept = FALSE)
  fit$changepoints
}

Fit a WERM changepoint model with squared loss and CV selection

Description

Selects the number of changepoints from ⁠{0, …, max_num_chgpts}⁠ by interleaved cross-validation, then refits on the full data.

Usage

werm_least_squares_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  use_base_loss_for_cv = FALSE,
  penalty = "none",
  alpha = 0
)
werm_least_squares_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  use_base_loss_for_cv = FALSE,
  penalty = "none",
  alpha = 0
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Numeric vector of length n. Ordered responses.

max_num_chgpts

Integer. Upper bound of the CV search grid.

delta

Integer. Minimum gap between candidate changepoints. Default 1.

search_method

Character. "efficient" (default) or "brute_force".

cv

Integer. Number of interleaved folds. Default 5.

fit_intercept

Logical. Default TRUE.

use_base_loss_for_cv

Logical. If TRUE, uses squared loss for CV segment fits and scoring instead of the default absolute error. Default FALSE.

penalty

Character. Passed to the inner fixed model. Default "none".

alpha

Numeric. Penalty strength. Default 0.

Value

An object of class c("werm_least_squares_cv", "werm_fit") with all elements from werm_least_squares() plus:

best_num_chgpts: Integer. CV-selected number of changepoints.
best_score: Numeric. Mean held-out score for the best model.
cv_results: Data frame with columns num_chgpts and mean_test_score.
num_chgpts_grid: Integer vector. Full CV search grid.
segment_bounds: List of integer pairs ⁠[start, stop]⁠ (1-indexed, half-open).
segment_coefs: Matrix ⁠(num_signals x p)⁠.
segment_intercepts: Numeric vector or NULL.

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLeastSquaresCV")) {
  set.seed(10)
  n <- 30L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:10, ] %*% c(3.5, 0),
    X[11:20, ] %*% c(-3.5, 0),
    X[21:30, ] %*% c(3.5, 0)
  ) + rnorm(n, sd = 0.05)
  fit <- werm_least_squares_cv(X, y, max_num_chgpts = 2L, cv = 3L,
                               delta = 3L, fit_intercept = FALSE)
  fit$best_num_chgpts
  fit$changepoints
  fit$cv_results
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLeastSquaresCV")) {
  set.seed(10)
  n <- 30L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  y <- c(
    X[1:10, ] %*% c(3.5, 0),
    X[11:20, ] %*% c(-3.5, 0),
    X[21:30, ] %*% c(3.5, 0)
  ) + rnorm(n, sd = 0.05)
  fit <- werm_least_squares_cv(X, y, max_num_chgpts = 2L, cv = 3L,
                               delta = 3L, fit_intercept = FALSE)
  fit$best_num_chgpts
  fit$changepoints
  fit$cv_results
}

Fit a WERM changepoint model with binary logistic loss

Description

Detects num_chgpts changepoints in ordered binary classification data.

Usage

werm_logistic(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  max_iter = 100L,
  tol = 1e-05,
  penalty = "l2",
  alpha = 1
)
werm_logistic(
  X,
  y,
  num_chgpts,
  delta = 1L,
  search_method = "efficient",
  fit_intercept = TRUE,
  max_iter = 100L,
  tol = 1e-05,
  penalty = "l2",
  alpha = 1
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Integer or factor vector of binary labels (two unique values).

num_chgpts

Integer. Number of changepoints to detect.

delta

Integer. Minimum gap between candidate changepoints during search. Default 1. Rule of thumb: max(1L, as.integer(0.05 * nrow(X))).

search_method

Character. "efficient" (default) or "brute_force".

fit_intercept

Logical. Whether each segment model includes an intercept. Default TRUE.

max_iter

Integer. Maximum L-BFGS-B iterations. Default 100.

tol

Numeric. Gradient-norm tolerance. Default 1e-5.

penalty

Character. Default "l2" (prevents divergence on separable segments). Set "none" only when segments are not separable.

alpha

Numeric. Default 1.0.

Value

An object of class c("werm_logistic", "werm_fit"). Contains all elements from werm_least_squares() plus:

classes: Character vector of length 2. The two class labels in sorted order. classes[2] is the positive class.

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLogistic")) {
  set.seed(3)
  n <- 30L; p <- 2L; true_cp <- 15L
  X <- matrix(rnorm(n * p), n, p)
  eta <- c(
    X[1:true_cp, ] %*% c(3, -3),
    X[(true_cp + 1L):n, ] %*% c(-3, 3)
  )
  y <- rbinom(n, 1L, 1 / (1 + exp(-eta)))
  fit <- werm_logistic(X, y, num_chgpts = 1L, delta = 3L,
                       fit_intercept = FALSE, max_iter = 100L)
  fit$changepoints
  fit$classes
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLogistic")) {
  set.seed(3)
  n <- 30L; p <- 2L; true_cp <- 15L
  X <- matrix(rnorm(n * p), n, p)
  eta <- c(
    X[1:true_cp, ] %*% c(3, -3),
    X[(true_cp + 1L):n, ] %*% c(-3, 3)
  )
  y <- rbinom(n, 1L, 1 / (1 + exp(-eta)))
  fit <- werm_logistic(X, y, num_chgpts = 1L, delta = 3L,
                       fit_intercept = FALSE, max_iter = 100L)
  fit$changepoints
  fit$classes
}

Fit a WERM changepoint model with logistic loss and CV selection

Description

Fit a WERM changepoint model with logistic loss and CV selection

Usage

werm_logistic_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  max_iter = 100L,
  tol = 1e-05,
  use_base_loss_for_cv = FALSE,
  penalty = "l2",
  alpha = 1
)
werm_logistic_cv(
  X,
  y,
  max_num_chgpts,
  delta = 1L,
  search_method = "efficient",
  cv = 5L,
  fit_intercept = TRUE,
  max_iter = 100L,
  tol = 1e-05,
  use_base_loss_for_cv = FALSE,
  penalty = "l2",
  alpha = 1
)

Arguments

X

Numeric matrix of shape ⁠(n, p)⁠. Observations must be in order.

y

Integer or factor vector of binary labels.

max_num_chgpts

Integer. Upper bound of the CV search grid.

delta

Integer. Minimum gap between candidate changepoints. Default 1.

search_method

Character. "efficient" (default) or "brute_force".

cv

Integer. Number of interleaved folds. Default 5.

fit_intercept

Logical. Default TRUE.

max_iter

Integer. Maximum L-BFGS-B iterations. Default 100.

tol

Numeric. Gradient-norm tolerance. Default 1e-5.

use_base_loss_for_cv

Logical. If TRUE, uses logistic loss for CV segment scoring instead of the default absolute error. Default FALSE.

penalty

Character. Default "l2".

alpha

Numeric. Default 1.0.

Value

An object of class c("werm_logistic_cv", "werm_fit"). Contains all CV elements plus classes (character vector of length 2).

Examples

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLogisticCV")) {
  set.seed(12)
  n <- 30L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  eta <- c(
    X[1:10, ] %*% c(3, -3),
    X[11:20, ] %*% c(-3, 3),
    X[21:30, ] %*% c(3, -3)
  )
  y <- rbinom(n, 1L, 1 / (1 + exp(-eta)))
  fit <- werm_logistic_cv(X, y, max_num_chgpts = 2L, cv = 3L, delta = 3L,
                          fit_intercept = FALSE, max_iter = 100L)
  fit$best_num_chgpts
  fit$changepoints
}

# Limit BLAS/OpenMP threads so example CPU time stays proportional to
# elapsed time on multicore CRAN machines.
Sys.setenv(
  OMP_NUM_THREADS = "1",
  OPENBLAS_NUM_THREADS = "1",
  MKL_NUM_THREADS = "1",
  BLAS_NUM_THREADS = "1"
)

if (nzchar(Sys.getenv("RETICULATE_PYTHON")) &&
    weightederm:::.weightederm_examples_available("WERMLogisticCV")) {
  set.seed(12)
  n <- 30L; p <- 2L
  X <- matrix(rnorm(n * p), n, p)
  eta <- c(
    X[1:10, ] %*% c(3, -3),
    X[11:20, ] %*% c(-3, 3),
    X[21:30, ] %*% c(3, -3)
  )
  y <- rbinom(n, 1L, 1 / (1 + exp(-eta)))
  fit <- werm_logistic_cv(X, y, max_num_chgpts = 2L, cv = 3L, delta = 3L,
                          fit_intercept = FALSE, max_iter = 100L)
  fit$best_num_chgpts
  fit$changepoints
}

Package 'weightederm'

Help Index

Extract last-segment coefficients

Description

Usage

Arguments

Value

Predict using the last-segment refit

Description

Usage

Arguments

Value

Summarise a fitted WERM model

Description

Usage

Arguments

Value

Configure which Python environment weightederm uses

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with Huber loss

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with Huber loss and CV selection

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with squared loss

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with squared loss and CV selection

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with binary logistic loss

Description

Usage

Arguments

Value

Examples

Fit a WERM changepoint model with logistic loss and CV selection

Description

Usage

Arguments

Value

Examples