Type: Package
Title: Analyzing Censored Factor Models
Version: 0.6.0
Description: Provides generation and estimation of censored factor models for high-dimensional data with censored errors (normal, t, logistic). Includes Sparse Orthogonal Principal Components (SOPC), and evaluation metrics. Based on Guo G. (2023) <doi:10.1007/s00180-022-01270-z>.
License: MIT + file LICENSE
Encoding: UTF-8
Language: en-US
Depends: R (≥ 3.5.0)
Imports: stats, MASS, psych, matrixcalc, crch
Suggests: testthat (≥ 3.0.0), ggplot2
NeedsCompilation: no
Packaged: 2025-10-18 01:51:44 UTC; 86188
RoxygenNote: 7.3.3
Author: Guangbao Guo [aut, cre], Tong Meng [aut]
Maintainer: Guangbao Guo <ggb11111111@163.com>
Repository: CRAN
Date/Publication: 2025-10-22 18:50:10 UTC

Censored Factor Models Data Generation

Description

Generate multivariate data that follow a latent factor structure with censoring errors drawn from Normal, Student-t or Logistic distributions. Convenience wrapper around rcnorm, rct, and rclogis.

Usage

CFM(n, p, m, cens.dist = c("normal", "t", "logistic"), df = 5, seed = NULL)

Arguments

n

sample size (n \times 1 observations).

p

number of manifest variables.

m

number of latent factors.

cens.dist

censoring error distribution: "normal", "t", or "logistic".

df

degrees of freedom when cens.dist = "t".

seed

optional random seed for reproducibility.

Value

A named list with components:

data

numeric n \times p matrix of observations.

F

factor scores matrix (n \times m).

A

factor loadings matrix (p \times m).

D

unique variances diagonal matrix (p \times p).

Examples


set.seed(2025)
# Normal censoring
obj <- CFM(n = 200, p = 10, m = 3, cens.dist = "normal")
head(obj$data)

# t-censoring with 6 d.f.
obj <- CFM(n = 300, p = 12, m = 4, cens.dist = "t", df = 6)
psych::KMO(obj$data)

Censored Factor Analysis via Principal Component (FanPC, pure R)

Description

Censored Factor Analysis via Principal Component (FanPC, pure R)

Usage

FanPC.CFM(
  data,
  m,
  A = NULL,
  D = NULL,
  p = NULL,
  cens.dist = c("normal", "t", "logistic"),
  df = NULL,
  cens.method = c("winsorise", "em"),
  cens_prop = 0.01,
  surv.obj = NULL,
  ctrl = NULL,
  verbose = NULL
)

Arguments

data

Numeric matrix or data frame of dimension n \times p.

m

Number of factors (< p).

A

Optional true loading matrix, used only for error calculation.

D

Optional true unique-variance diagonal matrix, used only for error calculation.

p

Number of variables (deprecated; detected automatically).

cens.dist

Error distribution, reserved for future use.

df

Degrees of freedom, reserved for future use.

cens.method

Censoring handling method; currently only "winsorise" is implemented. Defaults to "winsorise".

cens_prop

Winsorisation proportion, default 0.01.

surv.obj

Reserved for future use.

ctrl

Reserved for future use.

verbose

Reserved for future use.

Value

AF

Estimated loading matrix, p × m.

DF

Estimated unique-variance diagonal matrix, p × p.

MSESigmaA

Mean squared error of loadings (if A is provided).

MSESigmaD

Mean squared error of unique variances (if D is provided).

LSigmaA

Relative error of loadings (if A is provided).

LSigmaD

Relative error of unique variances (if D is provided).

Examples


library(CFM)
obj <- CFM(n = 500, p = 10, m = 2, cens.dist = "normal")
res <- FanPC.CFM(obj$data, m = 2, A = obj$A, D = obj$D, cens.method = "winsorise")
print(res$MSESigmaA)

PC2 for censored factor models (Top-2 principal components, pure R)

Description

PC2 for censored factor models (Top-2 principal components, pure R)

Usage

PC2.CFM(
  data,
  m,
  A = NULL,
  D = NULL,
  p = NULL,
  cens.dist = c("normal", "t", "logistic"),
  df = NULL,
  cens.method = c("winsorise", "em"),
  cens_prop = 0.01,
  surv.obj = NULL,
  ctrl = NULL,
  verbose = NULL
)

Arguments

data

Numeric matrix or data frame of dimension n \times p.

m

Number of factors (< p).

A

Optional true loading matrix, used only for error calculation.

D

Optional true unique-variance diagonal matrix, used only for error calculation.

p

Number of variables (deprecated; detected automatically).

cens.dist

Error distribution, reserved for future use.

df

Degrees of freedom, reserved for future use.

cens.method

Censoring handling method; currently only "winsorise" is implemented. Defaults to "winsorise".

cens_prop

Winsorisation proportion, default 0.01.

surv.obj

Reserved for future use.

ctrl

Reserved for future use.

verbose

Reserved for future use.

Value

AF

Estimated loading matrix, p × 2.

DF

Estimated unique-variance diagonal matrix, p × p.

MSESigmaA

Mean squared error of loadings (if A is provided).

MSESigmaD

Mean squared error of unique variances (if D is provided).

LSigmaA

Relative error of loadings (if A is provided).

LSigmaD

Relative error of unique variances (if D is provided).

Examples


library(CFM)
obj <- CFM(n = 500, p = 12, m = 2, cens.dist = "normal")
res <- PC2.CFM(obj$data, A = obj$A, D = obj$D)
print(res$MSESigmaA)

PPC2 for censored factor models (Top-2 principal components, pure R)

Description

PPC2 for censored factor models (Top-2 principal components, pure R)

Usage

PPC2.CFM(
  data,
  m,
  A = NULL,
  D = NULL,
  p = NULL,
  cens.dist = c("normal", "t", "logistic"),
  df = NULL,
  cens.method = c("winsorise", "em"),
  cens_prop = 0.01,
  surv.obj = NULL,
  ctrl = NULL,
  verbose = NULL
)

Arguments

data

Numeric matrix or data frame of dimension n \times p.

m

Number of factors (< p).

A

Optional true loading matrix, used only for error calculation.

D

Optional true unique-variance diagonal matrix, used only for error calculation.

p

Number of variables (deprecated; detected automatically).

cens.dist

Error distribution, reserved for future use.

df

Degrees of freedom, reserved for future use.

cens.method

Censoring handling method; currently only "winsorise" is implemented. Defaults to "winsorise".

cens_prop

Winsorisation proportion, default 0.01.

surv.obj

Reserved for future use.

ctrl

Reserved for future use.

verbose

Reserved for future use.

Value

AF

Estimated loading matrix, p × 2.

DF

Estimated unique-variance diagonal matrix, p × p.

MSESigmaA

Mean squared error of loadings (if A is provided).

MSESigmaD

Mean squared error of unique variances (if D is provided).

LSigmaA

Relative error of loadings (if A is provided).

LSigmaD

Relative error of unique variances (if D is provided).

Examples


library(CFM)
obj <- CFM(n = 500, p = 12, m = 2, cens.dist = "normal")
res <- PPC2.CFM(obj$data, A = obj$A, D = obj$D, cens.method = "winsorise")
print(res$MSESigmaA)

Basic censored-factor data simulator

Description

Generates multivariate data that follow a latent factor structure with censored errors (Normal, Student-t or Logistic).

Usage

censored_factor_models(
  n,
  p,
  m,
  distribution = c("normal", "t", "logistic"),
  df = NULL,
  seed = NULL
)

Arguments

n

Sample size (> 0).

p

Number of observed variables (> 0).

m

Number of latent factors (< p).

distribution

Error distribution: "normal" (default), "t", "logistic".

df

Degrees of freedom when distribution = "t".

seed

Optional random seed.

Value

A list with components:

data

numeric n × p matrix of observations

loadings

p × m factor loadings matrix

uniqueness

p × p diagonal uniqueness matrix

KMO

KMO measure of sampling adequacy

Bartlett_p

p-value of Bartlett's test

distribution

error distribution used

seed

random seed

Examples


set.seed(2025)
obj <- censored_factor_models(200, 6, 2)
psych::KMO(obj$data)