Package 'tramnet' reference manual

Title:	Penalized Transformation Models
Description:	Partially penalized versions of specific transformation models implemented in package 'mlt'. Available models include a fully parametric version of the Cox model, other parametric survival models (Weibull, etc.), normal and transformed-normal (Box-Cox type) linear models, and continuous outcome logistic regression. Models can be specified via a formula interface. Hyperparameter tuning is facilitated through model-based optimization functionalities from package 'mlr3mbo'. The accompanying vignette describes the methodology used in 'tramnet' in detail. Transformation models and model-based optimization are described in Hothorn et al. (2019) <doi:10.1111/sjos.12291> and Bischl et al. (2017) <doi:10.48550/arXiv.1703.03373>, respectively.
Authors:	Lucas Kook [cre, aut], Balint Tamasi [ctb], Sandra Siegfried [ctb], Samuel Pawel [ctb], Torsten Hothorn [ctb] (ORCID: <https://orcid.org/0000-0001-8301-0471>)
Maintainer:	Lucas Kook <[email protected]>
License:	GPL-2
Version:	0.1-0
Built:	2026-10-17 10:27:06 UTC
Source:	https://codeberg.org/lucaskook/tramnet

coef method for class `"tramnet"`

Description

coef method for class "tramnet"

Usage

## S3 method for class 'tramnet'
coef(object, with_baseline = FALSE, tol = 1e-06, scaled = FALSE, ...)
## S3 method for class 'tramnet'
coef(object, with_baseline = FALSE, tol = 1e-06, scaled = FALSE, ...)

Arguments

object

object of class "tramnet"

with_baseline

If TRUE, also prints coefficients for the baseline transformation

tol

tolerance when an estimate should be considered 0 and not returned (default: 1e-6)

scaled

only for models fitted via the formula interface: If FALSE (default), shift parameters are returned on the original scale of the covariates, otherwise on the scale of the internally scaled covariates (on which the penalty acts)

...

Additional arguments to coef

Value

Numeric vector containing the model shift parameter estimates

Author(s)

Torsten Hothorn, Lucas Kook

coef method for class `"tramnet_Lm"`

Description

coef method for class "tramnet_Lm"

Usage

## S3 method for class 'tramnet_Lm'
coef(
  object,
  with_baseline = FALSE,
  tol = 1e-06,
  as.lm = FALSE,
  scaled = FALSE,
  ...
)
## S3 method for class 'tramnet_Lm'
coef(
  object,
  with_baseline = FALSE,
  tol = 1e-06,
  as.lm = FALSE,
  scaled = FALSE,
  ...
)

Arguments

object

object of class "tramnet_Lm"

with_baseline

If TRUE, also prints coefficients for the baseline transformation

tol

tolerance when an estimate should be considered 0 and not returned (default: 1e-6)

as.lm

If TRUE parameters are rescaled to the usual parametrization of lm

scaled

...

Additional arguments to coef

Value

Numeric vector containing the linear model shift parameter estimates

Author(s)

Torsten Hothorn, Lucas Kook

Examples


data(cars)
m0 <- Lm(dist ~ 1, data = cars)
x <- as.matrix(cars[, "speed", drop = FALSE])
mt <- tramnet(m0, x = x, alpha = 0, lambda = 0)
coef(mt)
coef(mt, with_baseline = TRUE)
coef(mt, as.lm = TRUE)
coef(lm(dist ~ speed, data = cars))


data(cars)
m0 <- Lm(dist ~ 1, data = cars)
x <- as.matrix(cars[, "speed", drop = FALSE])
mt <- tramnet(m0, x = x, alpha = 0, lambda = 0)
coef(mt)
coef(mt, with_baseline = TRUE)
coef(mt, as.lm = TRUE)
coef(lm(dist ~ speed, data = cars))

Regularised Transformation Models via Formulae

Description

Penalized versions of the transformation models implemented in package tram, specified via a formula. These are convenience wrappers around tramnet, analogous to the corresponding functions Coxph, Colr, Lm, BoxCox, Lehmann and Survreg.

Usage

Coxphnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Colrnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Lmnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

BoxCoxnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Lehmannnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Survregnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)
Coxphnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Colrnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Lmnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

BoxCoxnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Lehmannnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Survregnet(
  formula,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Arguments

formula

an object of class "formula", see tram

data

an optional data frame containing the variables in the model

subset

an optional vector specifying a subset of observations to be used in the fitting process

weights

an optional vector of case weights to be used in the fitting process

offset

an optional vector of offsets

na.action

a function which indicates what should happen when the data contain NAs

lambda

a positive penalty parameter for the whole penalty function

alpha

a mixing parameter (between zero and one) defining the fraction between absolute and quadratic penalty terms

constraints

an optional list containing a matrix of linear inequality contraints on the regression coefficients (on the original scale of the covariates) and a vector specifying the rhs of the inequality

...

additional arguments to the corresponding function in package tram, e.g., order, log_first or support

Details

The left-hand side of formula (including strata, as in y | s ~ x) defines the unpenalized baseline transformation, all terms on the right-hand side are penalized. The design matrix is obtained via model.matrix without intercept and its columns are scaled to unit variance. The scaling of the training data is stored and re-used whenever newdata is supplied to one of the methods, or in cvl_tramnet. Coefficients are reported on the original scale of the covariates, see coef.tramnet.

Not all combinations of error distribution and censoring type lead to objective functions that can be solved by CVXR; see the table in the tramnet vignette.

Value

An object of class "tramnet", see tramnet

Author(s)

Lucas Kook

Examples


if (require("survival") & require("TH.data")) {
  data("GBSG2", package = "TH.data")
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  fit <- Coxphnet(surv ~ horTh + age + tsize + pnodes, data = GBSG2,
                  lambda = 5, alpha = 1, log_first = TRUE)
  coef(fit)
  coef(fit, scaled = TRUE)
  predict(fit, newdata = GBSG2[1:5, ], type = "survivor", q = 1000)
}
data("cars")
fit <- Lmnet(dist ~ speed, data = cars, lambda = 1, alpha = 0)
coef(fit, as.lm = TRUE)

if (require("survival") & require("TH.data")) {
  data("GBSG2", package = "TH.data")
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  fit <- Coxphnet(surv ~ horTh + age + tsize + pnodes, data = GBSG2,
                  lambda = 5, alpha = 1, log_first = TRUE)
  coef(fit)
  coef(fit, scaled = TRUE)
  predict(fit, newdata = GBSG2[1:5, ], type = "survivor", q = 1000)
}
data("cars")
fit <- Lmnet(dist ~ speed, data = cars, lambda = 1, alpha = 0)
coef(fit, as.lm = TRUE)

Cross validation for `"tramnet"` models

Description

k-fold cross validation for "tramnet" objects over a grid of the tuning parameters based on out-of-sample log-likelihood. For models fitted via the formula interface, the covariates are scaled within each training fold and the test fold is scaled accordingly.

Usage

cvl_tramnet(
  object,
  fold = 2,
  lambda = 0,
  alpha = 0,
  folds = NULL,
  fit_opt = FALSE
)
cvl_tramnet(
  object,
  fold = 2,
  lambda = 0,
  alpha = 0,
  folds = NULL,
  fit_opt = FALSE
)

Arguments

object

object of class "tramnet"

fold

number of folds for cross validation

lambda

values for lambda to iterate over

alpha

values for alpha to iterate over

folds

manually specify folds for comparison with other methods

fit_opt

If TRUE, returns the full model evaluated at optimal hyper parameters

Value

Returns out-of-sample logLik and coefficient estimates for corresponding folds and values of the hyperparameters as an object of class "cvl_tramnet"

Author(s)

Lucas Kook

Examples


set.seed(241068)
if (require("survival") & require("TH.data")) {
  data("GBSG2", package = "TH.data")
  X <- 1 * matrix(GBSG2$horTh == "yes", ncol = 1)
  colnames(X) <- "horThyes"
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  m <- Coxph(surv ~ 1, data = GBSG2, log_first = TRUE)
  mt <- tramnet(model = m, x = X, lambda = 0, alpha = 0)
  mc <- Coxph(surv ~ horTh, data = GBSG2)
  cvl_tramnet(mt, fold = 2, lambda = c(0, 1), alpha = c(0, 1))
  ## formula interface
  mf <- Coxphnet(surv ~ horTh + age, data = GBSG2, lambda = 0, alpha = 0,
                 log_first = TRUE)
  cvl_tramnet(mf, fold = 2, lambda = c(0, 1), alpha = c(0, 1))
}

set.seed(241068)
if (require("survival") & require("TH.data")) {
  data("GBSG2", package = "TH.data")
  X <- 1 * matrix(GBSG2$horTh == "yes", ncol = 1)
  colnames(X) <- "horThyes"
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  m <- Coxph(surv ~ 1, data = GBSG2, log_first = TRUE)
  mt <- tramnet(model = m, x = X, lambda = 0, alpha = 0)
  mc <- Coxph(surv ~ horTh, data = GBSG2)
  cvl_tramnet(mt, fold = 2, lambda = c(0, 1), alpha = c(0, 1))
  ## formula interface
  mf <- Coxphnet(surv ~ horTh + age, data = GBSG2, lambda = 0, alpha = 0,
                 log_first = TRUE)
  cvl_tramnet(mf, fold = 2, lambda = c(0, 1), alpha = c(0, 1))
}

Elastic net objective function for model based optimization

Description

This function generates an objective function for model-based optimization based on the cross-validated log-likelihood of a tramnet model with an elastic net penalty. It is not intended to be called by the user directly, instead it will be given as an argument to mbo_tramnet.

Usage

elnet_obj(
  object,
  minlambda = 0,
  maxlambda = 16,
  minalpha = 0,
  maxalpha = 1,
  folds,
  noisy = FALSE,
  fold
)
elnet_obj(
  object,
  minlambda = 0,
  maxlambda = 16,
  minalpha = 0,
  maxalpha = 1,
  folds,
  noisy = FALSE,
  fold
)

Arguments

object

object of class tramnet

minlambda

minimum value for lambda (default: 0)

maxlambda

maximum value for lambda (default: 16)

minalpha

minimum value for alpha (default: 0)

maxalpha

maximum value for alpha (default: 1)

folds

self specified folds for cross validation (mainly for reproducibility and comparability purposes)

noisy

indicates whether folds for k-fold cross-validation should be random for each iteration, leading to a noisy objective function (default: FALSE)

fold

fold for cross validation

Value

Single objective function for model based optimization of class ObjectiveRFun.

estfun method for class `"tramnet"`

Description

estfun method for class "tramnet" which computes the score contributions w.r.t. each model parameter.

Usage

## S3 method for class 'tramnet'
estfun(
  x,
  parm = coef(x, with_baseline = TRUE, tol = 0, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)
## S3 method for class 'tramnet'
estfun(
  x,
  parm = coef(x, with_baseline = TRUE, tol = 0, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)

Arguments

x

object of class "tramnet"

parm

parameters for evaluating the score

w

weights

newdata

data on which to compute the score contributions

...

additional arguments to estfun

Value

Matrix of score contributions w.r.t. model parameters evaluated at parm

Author(s)

Lucas Kook

Lasso objective function for model based optimization

Description

This function generates an objective function for model-based optimization based on the cross-validated log-likelihood of a tramnet model with a lasso penalty only. It is not intended to be called by the user directly, instead it will be given as an argument to mbo_tramnet.

Usage

lasso_obj(object, minlambda = 0, maxlambda = 16, folds, noisy = FALSE, fold)
lasso_obj(object, minlambda = 0, maxlambda = 16, folds, noisy = FALSE, fold)

Arguments

object

object of class tramnet

minlambda

minimum value for lambda (default: 0)

maxlambda

maximum value for lambda (default: 16)

folds

self specified folds for cross validation (mainly for reproducibility and comparability purposes)

noisy

indicates whether folds for k-fold cross-validation should be random for each iteration, leading to a noisy objective function (default: FALSE)

fold

fold for cross validation

Value

Single objective function for model based optimization of class ObjectiveRFun.

logLik method for class `"tramnet"`

Description

logLik method for class "tramnet"

Usage

## S3 method for class 'tramnet'
logLik(
  object,
  parm = coef(object, tol = 0, with_baseline = TRUE, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)
## S3 method for class 'tramnet'
logLik(
  object,
  parm = coef(object, tol = 0, with_baseline = TRUE, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)

Arguments

object

object of class "tramnet"

parm

parameters to evaluate the log likelihood at (for models fitted via the formula interface, shift parameters refer to the scaled covariates)

w

weights, defaults to the weights used for fitting object if newdata is missing

newdata

data to evaluate the log likelihood at

...

Additional arguments to logLik

Value

returns potentially weighted (w) log-likelihood based on object evaluated at parameters parm and data newdata. If newdata is missing, the penalized log-likelihood of the training data is returned, otherwise the (unpenalized) out-of-sample log-likelihood

Author(s)

Lucas Kook, Torsten Hothorn

Fit recommended regularized tram based on model based optimization output

Description

Extracts the "optimal" tuning parameters from an object returned by mbo_tramnet and fits the corresponding tramnet model

Usage

mbo_recommended(mbo_obj, m0, x, ...)
mbo_recommended(mbo_obj, m0, x, ...)

Arguments

mbo_obj

object return by mbo_tramnet

m0

null model of class "tram" or a fitted model of class "tramnet", which is refitted at the recommended tuning parameters (in this case, x and ... are ignored)

x

matrix of covariables

...

additional arguments to tramnet()

Value

Object of class "tramnet"

Model based optimization for regularized transformation models

Description

Uses model based optimization to find the optimal tuning parameter(s) in a regularized transformation model based on cross-validated log-likelihoods. Here the tramnet package makes use of the mlr3mbo interface for Bayesian Optimization in machine learning problems to maximize the cv-logLik as a black-box function of the tuning parameters alpha and lambda.

Usage

mbo_tramnet(
  object,
  fold = 2,
  n_design = 5,
  n_iter = 5,
  minlambda = 0,
  maxlambda = 16,
  minalpha = 0,
  maxalpha = 1,
  folds = NULL,
  learner = "regr.km",
  pred.type = "se",
  opt_crit = "ei",
  noisy = FALSE,
  obj_type = c("lasso", "ridge", "elnet"),
  verbose = TRUE,
  ...
)
mbo_tramnet(
  object,
  fold = 2,
  n_design = 5,
  n_iter = 5,
  minlambda = 0,
  maxlambda = 16,
  minalpha = 0,
  maxalpha = 1,
  folds = NULL,
  learner = "regr.km",
  pred.type = "se",
  opt_crit = "ei",
  noisy = FALSE,
  obj_type = c("lasso", "ridge", "elnet"),
  verbose = TRUE,
  ...
)

Arguments

object

object of class tramnet

fold

fold for cross validation

n_design

number of rows of the initial design matrix based on a random latin hypercube design

n_iter

number of iterations in the model based optimization procedure

minlambda

minimum value for lambda (default: 0)

maxlambda

maximum value for lambda (default: 16)

minalpha

minimum value for alpha (default: 0)

maxalpha

maximum value for alpha (default: 1)

folds

self specified folds for cross validation (mainly for reproducibility and comparability purposes)

learner

type of learner used as surrogate model in the optimization, either the key of a regression learner in mlr_learners (default: "regr.km", a Kriging model, which requires package DiceKriging) or an object of class LearnerRegr

pred.type

prediction type of the learner (default: "se")

opt_crit

optimization criterion, either the key of an acquisition function in mlr_acqfunctions or an object of class AcqFunction, default: expected improvement ("ei")

noisy

indicates whether folds for k-fold cross-validation should be random for each iteration, leading to a noisy objective function (default: FALSE)

obj_type

objective type, one of "lasso", "ridge" or "elnet"

verbose

toggle for a verbose output (default: TRUE)

...

additional arguments are ignored

Value

returns a list containing the recommended tuning parameters (x), the corresponding objective value (y), i.e., minus two times the cross-validated log-likelihood, the type of objective (obj_type) and the optimization instance (instance) of class OptimInstanceBatchSingleCrit, which contains the full archive of evaluated tuning parameters

Author(s)

Lucas Kook

Examples


if (require("survival") & require("TH.data") & require("DiceKriging")) {
  set.seed(241068)
  data("GBSG2", package = "TH.data")
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  mt <- Coxphnet(surv ~ horTh + age + tsize + pnodes, data = GBSG2,
                 lambda = 0, alpha = 0, log_first = TRUE)
  tmbo <- mbo_tramnet(mt, obj_type = "lasso", fold = 2, n_iter = 3,
                      verbose = FALSE)
  tmbo$x
  coef(mbo_recommended(tmbo, mt))
}


if (require("survival") & require("TH.data") & require("DiceKriging")) {
  set.seed(241068)
  data("GBSG2", package = "TH.data")
  GBSG2$surv <- with(GBSG2, Surv(time, cens))
  mt <- Coxphnet(surv ~ horTh + age + tsize + pnodes, data = GBSG2,
                 lambda = 0, alpha = 0, log_first = TRUE)
  tmbo <- mbo_tramnet(mt, obj_type = "lasso", fold = 2, n_iter = 3,
                      verbose = FALSE)
  tmbo$x
  coef(mbo_recommended(tmbo, mt))
}

model.matrix method for class `"tramnet"`

Description

model.matrix method for class "tramnet", returning the (scaled) design matrix of the penalized covariates

Usage

## S3 method for class 'tramnet'
model.matrix(object, newdata, ...)
## S3 method for class 'tramnet'
model.matrix(object, newdata, ...)

Arguments

object

object of class "tramnet"

newdata

optional data frame. For models fitted via the formula interface, the design matrix is computed from newdata and scaled as the training data. Not available for models fitted via the default interface

...

additional arguments are ignored

Value

Numeric matrix of covariates as used for fitting object (default) or computed from newdata

Author(s)

Lucas Kook

Plot regularization paths for `"prof_*"` classes

Description

Plot regularization paths and optionally log-likelihood trajectories of objects of class "prof_alpha" and "prof_lambda". Coefficient names are automatically added to the plot.

Usage

plot_path(object, plot_logLik = FALSE, ...)
plot_path(object, plot_logLik = FALSE, ...)

Arguments

object

object of class "prof_alpha" or "prof_lambda"

plot_logLik

Whether logLik trajectory should be plotted (default: FALSE)

...

additional arguments to plot

Value

None

Author(s)

Lucas Kook

Examples


library("tramnet")
if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit1 <- tramnet(y, x, lambda = 0, alpha = 1)
  pfl <- prof_lambda(fit1)
  plot_path(pfl)
  fit2 <- tramnet(y, x, lambda = 1, alpha = 1)
  pfa <- prof_alpha(fit2)
  plot_path(pfa)
}

library("tramnet")
if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit1 <- tramnet(y, x, lambda = 0, alpha = 1)
  pfl <- prof_lambda(fit1)
  plot_path(pfl)
  fit2 <- tramnet(y, x, lambda = 1, alpha = 1)
  pfa <- prof_alpha(fit2)
  plot_path(pfa)
}

plot method for class `"tramnet"`

Description

plot method for class "tramnet"

Usage

## S3 method for class 'tramnet'
plot(
  x,
  newdata,
  type = c("distribution", "survivor", "density", "logdensity", "hazard", "loghazard",
    "cumhazard", "quantile", "trafo"),
  q = NULL,
  prob = 1:(K - 1)/K,
  K = 50,
  col = rgb(0.1, 0.1, 0.1, 0.1),
  lty = 1,
  add = FALSE,
  ...
)
## S3 method for class 'tramnet'
plot(
  x,
  newdata,
  type = c("distribution", "survivor", "density", "logdensity", "hazard", "loghazard",
    "cumhazard", "quantile", "trafo"),
  q = NULL,
  prob = 1:(K - 1)/K,
  K = 50,
  col = rgb(0.1, 0.1, 0.1, 0.1),
  lty = 1,
  add = FALSE,
  ...
)

Arguments

x

object of class "tramnet"

newdata

data used to predict and plot

type

type of plot produced

q

vector of quantiles

prob

vector of probabilities

K

number of data points to plot

col

see plot

lty

see plot

add

see plot

...

additional options to plot

Value

None

Author(s)

Lucas Kook

predict method for class `"tramnet"`

Description

predict method for class "tramnet"

Usage

## S3 method for class 'tramnet'
predict(object, newdata, ...)
## S3 method for class 'tramnet'
predict(object, newdata, ...)

Arguments

object

object of class "tramnet"

newdata

data used for prediction, defaults to the data used for fitting object. For models fitted via the formula interface, newdata contains the original (unscaled) covariates

...

Additional arguments to predict.ctm

Value

Vector of predictions based on object evaluated at each row of newdata

Author(s)

Lucas Kook

print summary method for class `"tramnet"`

Description

print summary method for class "tramnet"

Usage

## S3 method for class 'summary.tramnet'
print(x, digits = max(3L, getOption("digits") - 3L), ...)
## S3 method for class 'summary.tramnet'
print(x, digits = max(3L, getOption("digits") - 3L), ...)

Arguments

x

object of class "tramnet"

digits

number of digits to print

...

additional arguments

Value

prints textual summary in the console and returns an invisible copy of the "tramnet" object

Author(s)

Lucas Kook

print method for class `"tramnet"`

Description

print method for class "tramnet"

Usage

## S3 method for class 'tramnet'
print(x, ...)
## S3 method for class 'tramnet'
print(x, ...)

Arguments

x

object of class "tramnet"

...

additional arguments to summary

Value

prints textual summary in the console and returns an invisible copy of the "tramnet" object

Author(s)

Lucas Kook

Profiling tuning parameters

Description

Computes the regularization path of all coefficients for a single tuning, alpha, parameter over a sequence of values.

Usage

prof_alpha(model, min_alpha = 0, max_alpha = 1, nprof = 5, as.lm = FALSE)
prof_alpha(model, min_alpha = 0, max_alpha = 1, nprof = 5, as.lm = FALSE)

Arguments

model

model of class tramnet

min_alpha

minimal value of alpha (default = 0)

max_alpha

maximal value of alpha (default = 15)

nprof

number of profiling steps (default = 5)

as.lm

return scaled coefficients for class "tramnet_Lm"

Value

Object of class "prof_alpha" which contains the regularization path of all coefficients and the log-likelihood over the mixing parameter alpha

Author(s)

Lucas Kook

Examples


library("tramnet")

if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit <- tramnet(y, x, lambda = 1, alpha = 1)
  pfa <- prof_alpha(fit)
  plot_path(pfa)
}

library("tramnet")

if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit <- tramnet(y, x, lambda = 1, alpha = 1)
  pfa <- prof_alpha(fit)
  plot_path(pfa)
}

Profiling tuning parameters

Description

Computes the regularization path of all coefficients for a single tuning parameter, lambda, over a sequence of values.

Usage

prof_lambda(model, min_lambda = 0, max_lambda = 15, nprof = 5, as.lm = FALSE)
prof_lambda(model, min_lambda = 0, max_lambda = 15, nprof = 5, as.lm = FALSE)

Arguments

model

model of class "tramnet"

min_lambda

minimal value of lambda (default = 0)

max_lambda

maximal value of lambda (default = 15)

nprof

number of profiling steps (default = 5)

as.lm

return scaled coefficients for class "tramnet_Lm"

Value

Object of class "prof_lambda" which contains the regularization path of all coefficients and the log-likelihood over the penalty parameter lambda

Author(s)

Lucas Kook

Examples


library("tramnet")
if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit <- tramnet(y, x, lambda = 0, alpha = 1)
  pfl <- prof_lambda(fit)
  plot_path(pfl)
}

library("tramnet")
if (require("survival") & require("penalized")) {
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70))
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit <- tramnet(y, x, lambda = 0, alpha = 1)
  pfl <- prof_lambda(fit)
  plot_path(pfl)
}

residuals method for class `"tramnet"`

Description

residuals method for class "tramnet"

Usage

## S3 method for class 'tramnet'
residuals(
  object,
  parm = coef(object, tol = 0, with_baseline = TRUE, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)
## S3 method for class 'tramnet'
residuals(
  object,
  parm = coef(object, tol = 0, with_baseline = TRUE, scaled = TRUE),
  w = NULL,
  newdata,
  ...
)

Arguments

object

object of class "tramnet"

parm

parameters to evaluate score at

w

weights

newdata

data to evaluate score at

...

additional arguments to residuals

Value

Returns a numeric vector of residuals for each row in newdata

Author(s)

Lucas Kook

Ridge objective function for model based optimization

Description

This function generates an objective function for model-based optimization based on the cross-validated log-likelihood of a tramnet model with a ridge penalty only. It is not intended to be called by the user directly, instead it will be given as an argument to mbo_tramnet.

Usage

ridge_obj(object, minlambda = 0, maxlambda = 16, folds, noisy = FALSE, fold)
ridge_obj(object, minlambda = 0, maxlambda = 16, folds, noisy = FALSE, fold)

Arguments

object

object of class tramnet

minlambda

minimum value for lambda (default: 0)

maxlambda

maximum value for lambda (default: 16)

folds

self specified folds for cross validation (mainly for reproducibility and comparability purposes)

noisy

indicates whether folds for k-fold cross-validation should be random for each iteration, leading to a noisy objective function (default: FALSE)

fold

fold for cross validation

Value

Single objective function for model based optimization of class ObjectiveRFun.

simulate method for class `"tramnet"`

Description

simulate method for class "tramnet"

Usage

## S3 method for class 'tramnet'
simulate(object, nsim = 1, seed = NULL, newdata, bysim = TRUE, ...)
## S3 method for class 'tramnet'
simulate(object, nsim = 1, seed = NULL, newdata, bysim = TRUE, ...)

Arguments

object

object of class "tramnet"

nsim

number of simulation

seed

random number generator seed

newdata

new data to simulate from, defaults to the data used for fitting object

bysim

see simulate.ctm

...

Additional arguments to simulate.ctm

Value

returns a list of data.frames containing parametric bootstrap samples based on the data supplied in newdata

Author(s)

Lucas Kook

summary method for class `"tramnet"`

Description

summary method for class "tramnet"

Usage

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

Arguments

object

object of class "tramnet"

...

additional arguments

Value

Returns an object of class "summary.tramnet" containing information about the model, optimization status, sparsity and tuning parameters

Author(s)

Lucas Kook

Regularised Transformation Models

Description

Partially penalized and constrained transformation models, including Cox models and continuous outcome logistic regression. The methodology is described in the tramnet vignette accompanying this package.

Usage

tramnet(model, ...)

## Default S3 method:
tramnet(model, x, lambda, alpha, constraints = NULL, ...)

## S3 method for class 'formula'
tramnet(
  model,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)
tramnet(model, ...)

## Default S3 method:
tramnet(model, x, lambda, alpha, constraints = NULL, ...)

## S3 method for class 'formula'
tramnet(
  model,
  data,
  subset,
  weights,
  offset,
  na.action = na.omit,
  lambda,
  alpha,
  constraints = NULL,
  ...
)

Arguments

model

an object of class "tram" as returned by any of the modelling functions from package tram (default method) or a formula describing the model (formula method)

...

additional parameters to psolve (default method) or to tram (formula method), e.g., distribution, transformation, order or log_first

x

a numeric matrix, where each row corresponds to the same row in the data argument used to fit model

lambda

a positive penalty parameter for the whole penalty function

alpha

a mixing parameter (between zero and one) defining the fraction between absolute and quadratic penalty terms

constraints

an optional list containing a matrix of linear inequality contraints on the regression coefficients and a vector specifying the rhs of the inequality. In the formula method, the constraints refer to the coefficients on the original scale of the covariates

data

an optional data frame containing the variables in the model

subset

an optional vector specifying a subset of observations to be used in the fitting process

weights

an optional vector of case weights to be used in the fitting process

offset

an optional vector of offsets

na.action

a function which indicates what should happen when the data contain NAs

Details

The default method takes an unpenalized baseline model of class "tram", for instance Coxph(y ~ 1, data = d), and a matrix of covariates x, which should be scaled by the user. The formula method fits the baseline model and sets up the design matrix in one go. It is the penalized analogue of tram: the model is fixed through the arguments distribution and transformation (and others) passed via ... to tram. Convenience wrappers such as Coxphnet exist for the most common models.

In the formula method, the left-hand side of model (including strata, as in y | s ~ x) defines the baseline transformation, all terms on the right-hand side are penalized. The design matrix is obtained via model.matrix without intercept and its columns are scaled to unit variance. They are not centered, since the baseline transformation absorbs any shift in location. The scaling of the training data is stored and re-used whenever newdata is supplied, e.g., in predict, logLik or cvl_tramnet. Coefficients are reported on the original scale of the covariates, see coef.tramnet.

Value

An object of class "tramnet" with coef, logLik, summary, simulate, residuals, predict, model.matrix and plot methods

Author(s)

Lucas Kook, Balint Tamasi, Sandra Sigfried

Examples


if (require("penalized") & require("survival")) {
  ## --- Comparison with penalized
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46,
                          data = nki70))
  fit <- penalized(response = resp, penalized = x, lambda1 = 1, lambda2 = 0,
                   standardize = FALSE, data = nki70)
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit2 <- tramnet(y, x, lambda = 1, alpha = 1) ## L1 only
  coef(fit)
  coef(fit2)

  ## --- Formula interface, scaling is done internally
  fit3 <- tramnet(resp ~ DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70,
                  lambda = 1, alpha = 1, distribution = "MinExtrVal",
                  transformation = "smooth", negative = FALSE, order = 10,
                  log_first = TRUE)
  coef(fit3)
  coef(fit3, scaled = TRUE)
}

if (require("penalized") & require("survival")) {
  ## --- Comparison with penalized
  data("nki70", package = "penalized")
  nki70$resp <- with(nki70, Surv(time, event))
  x <- scale(model.matrix( ~ 0 + DIAPH3 + NUSAP1 + TSPYL5 + C20orf46,
                          data = nki70))
  fit <- penalized(response = resp, penalized = x, lambda1 = 1, lambda2 = 0,
                   standardize = FALSE, data = nki70)
  y <- Coxph(resp ~ 1, data = nki70, order = 10, log_first = TRUE)
  fit2 <- tramnet(y, x, lambda = 1, alpha = 1) ## L1 only
  coef(fit)
  coef(fit2)

  ## --- Formula interface, scaling is done internally
  fit3 <- tramnet(resp ~ DIAPH3 + NUSAP1 + TSPYL5 + C20orf46, data = nki70,
                  lambda = 1, alpha = 1, distribution = "MinExtrVal",
                  transformation = "smooth", negative = FALSE, order = 10,
                  log_first = TRUE)
  coef(fit3)
  coef(fit3, scaled = TRUE)
}

Package 'tramnet'

Help Index

coef method for class "tramnet"

Description

Usage

Arguments

Value

Author(s)

coef method for class "tramnet_Lm"

Description

Usage

Arguments

Value

Author(s)

Examples

Regularised Transformation Models via Formulae

Description

Usage

Arguments

Details

Value

Author(s)

Examples

Cross validation for "tramnet" models

Description

Usage

Arguments

Value

Author(s)

Examples

Elastic net objective function for model based optimization

Description

Usage

Arguments

Value

estfun method for class "tramnet"

Description

Usage

Arguments

Value

Author(s)

Lasso objective function for model based optimization

Description

Usage

Arguments

Value

logLik method for class "tramnet"

Description

Usage

Arguments

Value

Author(s)

Fit recommended regularized tram based on model based optimization output

Description

Usage

Arguments

Value

Model based optimization for regularized transformation models

Description

Usage

Arguments

Value

Author(s)

Examples

model.matrix method for class "tramnet"

Description

Usage

Arguments

Value

Author(s)

Plot regularization paths for "prof_*" classes

Description

Usage

Arguments

Value

Author(s)

Examples

plot method for class "tramnet"

Description

Usage

coef method for class `"tramnet"`

coef method for class `"tramnet_Lm"`

Cross validation for `"tramnet"` models

estfun method for class `"tramnet"`

logLik method for class `"tramnet"`

model.matrix method for class `"tramnet"`

Plot regularization paths for `"prof_*"` classes

plot method for class `"tramnet"`

predict method for class `"tramnet"`

print summary method for class `"tramnet"`

print method for class `"tramnet"`

residuals method for class `"tramnet"`

simulate method for class `"tramnet"`

summary method for class `"tramnet"`