SLOPE 1.0.0
CRAN release: 2025-06-30
This update of SLOPE brings an entirely different C++ implementation of the underlying package based on the C++ library libslope. It comes with several large and breaking changes with respect to the previous version of the package.
We realized that this may throw off some users, and hope that you will be patient with dealing with the large number of breaking changes.
Breaking Changes
- The
caretSLOPE()
function that was deprecated has now been removed from the package. - Fields
unique
,violations
, andactive_sets
are no longer stored in theSLOPE
object. These fields were typically only used for debugging purposes. - The
prox_method
andmethod
arguments inSLOPE()
andsortedL1Prox()
, respectively, have been removed. The proximal operator is now always computed using the fast stack-based algorithm. There was never any reason to use the slower PAVA algorithm. - The ADMM solver has been removed from the package. Calling
SLOPE()
withsolver = "admm"
will now throws a warning and the value will be automatically set to"auto"
. -
alpha
is now scaled byn
(the number of observations) and differences with respect to the type of scaling are no longer taken into account. - The object
coefficients
fromSLOPE()
is now a list of sparse matrices (rather than a three-dimensional array as before). Now it contains only the coefficients and not the intercepts. The intercepts are instead stored inintercepts
in the returned object and are always present even ifintercept = FALSE
. - The behavior of
coef.SLOPE()
has changed somewhat, and ifsimplify = FALSE
, then the returned object is now instead a list of sparse matrices (rather than a three-dimensional array as before). - The default value of
q
inSLOPE()
has changed from0.1 * min(1, NROW(x) / NCOL(x))
to0.1
. - Arguments
sigma
,n_sigma
, andlambda_min_ratio
inSLOPE()
that were previously deprecated have been removed. -
SLOPE()
now internally solves the problem normalized by scaling with the number of observations, which means that values returned indeviance
andprmals
andduals
ifdiagnostics = TRUE
are now scaled byn
. -
path_length
inSLOPE()
now defaults to 100 (previously 20). -
tol_dev_ratio
inSLOPE()
now defaults to0.999
(previously0.995
). - Plots from
plot.SLOPE()
now use base R graphics rather than ggplot2. This means that the plots are more difficult to customize but plot much more faster when there are many variables and significantly reduces the dependency load of the package. For plots of trained SLOPE objects, which used to be faceted on theq
parameter, the user now needs to use the standard base R graphics API to facet plots viapar(mfrow = c(1, 2))
or similar.
Deprecated Functionality
- Arguments
tol_rel_gap
,tol_infeas
,tol_abs
,tol_rel
,tol_rel_coef
inSLOPE()
are now deprecated. The solvers now all rely on the same tolerance criterion, which is set bytol
and uses the duality gap normalized by the current primal value. - Arguments
screen
andscreen_alg
are now deprecated and have no effect. Feature screening is always used. These arguments were only used for debugging. - The argument
verbosity
inSLOPE()
is now defunct and has no effect. - The argument
prox_method
inSLOPE()
andsortedL1Prox()
is now defunct and has no effect.
New Features
- Centering
x
inSLOPE()
is now allowed again, even when the matrix is sparse. - Out-of-memory matrices are now allowed through the
bigmemory
package. Only support for dense matrices is available at the moment. - Centers and scales can now be specified manually by providing vectors to
center
andscale
inSLOPE()
. - A new solver based on a hybrid method of proximal gradient descent and coordinate descent is available and used by default by the Gaussian and binomial families. Use it by specifying
solver = "hybrid"
. - Solver can now be set to
"auto"
, in which case the package automatically chooses a solver. - The returned duality gaps when
diagnostics = TRUE
are now true duality gaps, computed by guaranteeing that the dual variable is feasible (which was not the case previously). -
scale
inSLOPE()
gains a new option"max_abs"
which scales the columns ofx
by their maximum absolute value. - When
alpha = "estimate"
, there is a now an iteration limit in case the algorithm does not converge to one set of features. Thanks @RomanParzer. -
plot.SLOPE()
gains a new argumentmagnitudes
, which causes the plot to only show the magnitudes of the coefficients (which helps if you want to visualize cluster structure). -
plot.SLOPE()
gains a new argumentadd_labels
, which add numbers for the coefficients to the plot. Set toFALSE
by default. - Relaxed SLOPE models can now be fit by specifying
gamma
inSLOPE()
. -
plot.trainedSLOPE()
gains a new argumentindex
, to select which of the hyperparameter combinations to plot for. - There’s a new function
plotClusters()
, which allows plotting the cluster structure in SLOPE. Thanks, 1! -
SLOPE()
gains a new argumentcd_type
, to control the type of coordinate descent used for the hybrid solver, with options"cyclical"
and"permuted"
.
Bug Fixes
- Return correct model when training for AUC in
trainSLOPE()
.
Performance Improvements
The new hybrid algorithm that’s implemented in libslope and now used in the package constitutes a major upgrade in terms of performance.
- The solver is now much more memory-efficient and can avoid copies of the design matrix entirely by normalizing the columns just-in-time. This is the standard behavior. Future versions of the package will allow the user to specify whether to copy (and modify) the design matrix or not.
Dependencies
We have made an effort to reduce the footprint of the package and reduce the number of dependencies.
- The package now relies on Eigen (through RcppEigen) rather than Armadillo, which means that there is no longer any reliance on BLAS and LAPACK libraries.
- The dependency on
ggplot2
is removed. - The
vdiffr
,tidyr
,dplyr
,bench
,scales
, andglmnet
packages in theSuggests
field that were used for testing are now removed.
SLOPE 0.5.0
CRAN release: 2022-06-09
Major changes
-
plot.SLOPE()
,plot.trainSLOPE()
andplotDiagnostics()
have been reimplemented in ggplot2.
SLOPE 0.4.0
CRAN release: 2021-12-10
New Functions
-
sortedL1Prox()
is a new function that computes the proximal operator for the sorted L1 norm (the penalty term in SLOPE). -
regularizationWeights()
is a new function that returns the penalty weights (lambda sequence) for SLOPE or OSCAR.
Major changes
- The parametrization for OSCAR models have been corrected and changed. As a result,
SLOPE()
gains two arguments:theta1
andtheta2
to control the behavior using the parametrization from L. W. Zhong and J. T. Kwok, “Efficient sparse modeling with automatic feature grouping,” IEEE Transactions on Neural Networks and Learning Systems, vol. 23, no. 9, pp. 1436–1447, Sep. 2012, doi: 10.1109/TNNLS.2012.2200262.q
is no longer used with OSCAR models. Thanks, Nuno Eusebio. -
SLOPE()
has gained a new argument,prox_method
, which allows the user to select prox algorithm to use. There is no an additional algorithm in the package, based on the PAVA algorithm used in isotonic regression, that can be used. Note that this addition is mostly of academic interest and does not need to be changed by the user.
Minor Changes
- The
q
parameter is no longer allowed to be smaller than1e-6
to avoid constructions of regularization paths with infinitelambda
values. - The
lambda
argument inSLOPE()
now also allowed the input"lasso"
to obtain the standard lasso. - The performance of
trainSLOPE()
Bug Fixes
- For very small numbers of observations (10 or so), the regularization weights for
lambda = "gaussian"
were incorrectly computed, increasing and then decreasing. This is now fixed and regularization weights in this case are now always non-increasing. - Misclassification error was previously computed incorrectly in
trainSLOPE()
for multinomial models (thanks @jakubkala and 1) - Performance of
trainSLOPE()
was previously hampered by erroneous refitting of the models, which has been fixed now (thanks @jakubkala and
Deprecated and Defunct
-
yvar
argument inplotDiagnostics()
that was previously deprecated is now defunct. - Using
missclass
for themeasure
argument intrainSLOPE()
has been deprecated in favor ofmisclass
.
SLOPE 0.3.2
CRAN release: 2020-07-10
Minor changes
- Added
tol_rel_coef_change
argument toSLOPE()
as a convergence criterion for the FISTA solver that sets a tolerance for the relative change in coefficients across iterations.
SLOPE 0.3.0
CRAN release: 2020-07-02
Major changes
- Scaling of
alpha
(previouslysigma
) is now invariant to the number of observations, which is achieved by scaling the penalty part of the objective by the square root of the number of observations ifscale = "l2"
and the number of observations ifscale = "sd"
or"none"
. No scaling is applied whenscale = "l1"
. - The
sigma
argument is deprecated in favor ofalpha
inSLOPE()
,coef.SLOPE()
, andpredict.SLOPE()
. - The
n_sigma
argument is deprecated in favor ofpath_length
inSLOPE()
- The
lambda_min_ratio
argument is deprecated in favor ofalpha_min_ratio
inSLOPE()
- The default for argument
lambda
inSLOPE()
has changed from"gaussian"
to"bh"
. - Functions and arguments deprecated in 0.2.0 are now defunct and have been removed from the package.
-
scale = "sd"
now scales with the population standard deviation rather than the sample standard deviation, i.e. the scaling factor now used is the number of observations (and not the number of observations minus one as before).
Minor changes
- Default
path_length
has changed from 100 to 20. -
plot.SLOPE()
has gained an argumentx_variable
that controls what is plotted on the x axis. - A warning is now thrown if the maximum number of passes was reached anywhere along the path (and prints where as well).
- If the
max_variables
criterion is hit, the solution path returned will now include also the last solution (which was not the case before). Thanks,
SLOPE 0.2.1
CRAN release: 2020-04-16
Minor changes
- A few examples in
deviance()
andSLOPE()
that were taking too long to execute have been removed or modified.
SLOPE 0.2.0
This version of SLOPE represents a major change to the package. We have merged functionality from the owl package into this package, which means there are several changes to the API, including deprecated functions.
Major changes
-
SLOPE_solver()
,SLOPE_solver_matlab()
,prox_sorted_L1()
, andcreate_lambda()
have been deprecated (and will be defunct in the next version of SLOPE) - arguments
X
,fdr
, andnormalize
have been deprecated inSLOPE()
and replaced byx
,q
,scale
andcenter
, respectively - options
"default"
and"matlab"
to argumentsolver
inSLOPE()
have been deprecated and replaced with"fista"
and"admm"
, which uses the accelerated proximal gradient method FISTA and alternating direction of multipliers method (ADMM) respectively - ADMM has been implemented as a solver for
family = "gaussian"
- binomial, poisson, and multinomial families are now supported (using
family
argument inSLOPE()
) - input to
lambda
is now scaled (divided by) the number of observations (rows) inx
- predictor screening rules have been implemented and are activated by calling
screen = TRUE
inSLOPE()
. The type of algorithm can also be set viascreen_alg
. -
SLOPE()
now returns an object of class"SLOPE"
(and an additional class depending on input tofamily
inSLOPE()
-
SLOPE
objects gaincoef()
andplot()
methods. -
SLOPE
now uses screening rules to speed up model fitting in the high-dimensional regime - most of the code is now written in C++ using the Rcpp and RcppArmadillo packages
- a new function
trainSLOPE()
trains SLOPE with repeated k-folds cross-validation - a new function
caretSLOPE()
enables model-tuning using the caret package -
SLOPE()
now fits an entire path of regularization sequences by default - the
normalize
option toSLOPE()
has been replaced byscale
andcenter
, which allows granular options for standardization - sparse matrices (from the Matrix package) can now be used as input
- there are now five datasets included in the package
- the introductory vignette has been replaced
Minor changes
- a new function
deviance()
returns the deviance from the fit - a new function
score()
can be used to assess model performance against new data - a new function
plotDiagnostics()
has been included to visualize data from the solver (ifdiagnostics = TRUE
in the call toSLOPE()
) - OSCAR-type penalty sequences can be used by setting
lambda = "oscar" in the call to
SLOPE()` - the test suite for the package has been greatly extended