Package 'PPSFS'

Partial Profile Score Feature Selection for GLMs

Description

ppsfs: PPSFS for main-effects.

ppsfsi: PPSFS for interaction effects.

Usage

ppsfs(
  x,
  y,
  family,
  keep = NULL,
  I0 = NULL,
  ...,
  ebicFlag = 1,
  maxK = min(NROW(x) - 1, NCOL(x) + length(I0)),
  verbose = FALSE
)

ppsfsi(
  x,
  y,
  family,
  keep = NULL,
  ...,
  ebicFlag = 1,
  maxK = min(NROW(x) - 1, choose(NCOL(x), 2)),
  verbose = FALSE
)

Arguments

x	Matrix.
y	Vector.
family	See glm and family.
keep	Initial set of features that are included in model fitting.
I0	Index set of interaction effects to be identified.
...	Additional parameters for glm.fit.
ebicFlag	The procedure stops when the EBIC increases after ebicFlag times.
maxK	Maximum number of identified features.
verbose	Print the procedure path?

Details

That ppsfs(x, y, family="gaussian") is an implementation to sequential lasso method proposed by Luo and Chen(2014, <doi:10/f6kfr6>).

Value

Index set of identified features.

References

Z. Xu, S. Luo and Z. Chen (2022). Partial profile score feature selection in high-dimensional generalized linear interaction models. Statistics and Its Interface. doi:10.4310/21-SII706

Examples

## ***************************************************
## Identify main-effect features
## ***************************************************
set.seed(2022)
n <- 300
p <- 1000
x <- matrix(rnorm(n*p), n)
eta <- drop( x[, 1:3] %*% runif(3, 1.0, 1.5) )
y <- eta + rnorm(n, sd=sd(eta)/5)
print( A <- ppsfs(x, y, 'gaussian', verbose=TRUE) )

## ***************************************************
## Identify interaction effects
## ***************************************************
set.seed(2022)
n <- 300
p <- 150
x <- matrix(rnorm(n*p), n)
eta <- drop( cbind(x[, 1:3], x[, 4:6]*x[, 7:9]) %*% runif(6, 1.0, 1.5) )
y <- eta + rnorm(n, sd=sd(eta)/5)
print( group <- ppsfsi(x, y, 'gaussian', verbose=TRUE) )
print( A <- ppsfs(x, y, "gaussian", I0=group, verbose=TRUE) )

print( A <- ppsfs(x, y, "gaussian", keep=c(1, "5:8"), 
                  I0=group, verbose=TRUE) )

---

Partial Profile Score Feature Selection for GLMs

Description

This is a simplified but more stable implementation for main-effects selection compared to ppsfs. It removes the standaridization, but adds the weights in the partial profile score.

Usage

ppsfs.fit(
  x,
  y,
  family,
  fitFun = glm.fit,
  ...,
  keep = NULL,
  maxK = NULL,
  verbose = FALSE
)

Arguments

x	Matrix.
y	Vector.
family	See glm and family.
fitFun	Fitting function, by default glm.fit.
...	Additional parameters for glm.fit.
keep	Initial index set of features that are included in model fitting.
maxK	Maximum number of identified features.
verbose	Print the feature path?

Details

That ppsfs(x, y, family="gaussian") is an implementation to sequential lasso method proposed by Luo and Chen(2014, doi:10/f6kfr6).

References

Z. Xu, S. Luo and Z. Chen (2022). Partial profile score feature selection in high-dimensional generalized linear interaction models. Statistics and Its Interface. doi:10.4310/21-SII706

Examples

set.seed(2025)
n <- 300
p <- 1000
x <- matrix(rnorm(n*p), n)
eta <- drop( x[, 1:3] %*% runif(3, 1.0, 1.5) )
y <- eta + rnorm(n, sd=sd(eta)/5)
print( A <- ppsfs.fit(x, y, 'gaussian', verbose=TRUE) )

---