SLOPE 1.0.0

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, and active_sets are no longer stored in the
  SLOPE object. These fields were typically only used for debugging purposes.
• The prox_method and method arguments in SLOPE() and sortedL1Prox(),
  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() with
  solver = "admm" will now throws a warning and the value will be
  automatically set to "auto".
• alpha is now scaled by n (the number of observations) and differences with
  respect to the type of scaling are no longer taken into account.
• The object coefficients from SLOPE() 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 in
  intercepts in the returned object and are always present even if
  intercept = FALSE.
• The behavior of coef.SLOPE() has changed somewhat, and if
  simplify = 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 in SLOPE() has changed from
  0.1 * min(1, NROW(x) / NCOL(x)) to 0.1.
• Arguments sigma, n_sigma, and lambda_min_ratio in SLOPE() 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 in deviance and
  prmals and duals if diagnostics = TRUE are now scaled by n.
• path_length in SLOPE() now defaults to 100 (previously 20).
• tol_dev_ratio in SLOPE() now defaults to 0.999 (previously 0.995).
• 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
  the q parameter, the user now needs to use the standard base R graphics API
  to facet plots via par(mfrow = c(1, 2)) or similar.

Deprecated Functionality

• Arguments tol_rel_gap, tol_infeas, tol_abs, tol_rel, tol_rel_coef in
  SLOPE() are now deprecated. The solvers now all rely on the same tolerance
  criterion, which is set by tol and uses the duality gap normalized by the
  current primal value.
• Arguments screen and screen_alg are now deprecated and have no effect.
  Feature screening is always used. These arguments were only used for
  debugging.
• The argument verbosity in SLOPE() is now defunct and has no effect.
• The argument prox_method in SLOPE() and sortedL1Prox() is now defunct
  and has no effect.

New Features

• Centering x in SLOPE() 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 and scale in SLOPE().
• 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 in SLOPE() gains a new option "max_abs" which scales the columns
  of x 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 argument magnitudes, 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 argument add_labels, which add numbers for the
  coefficients to the plot. Set to FALSE by default.
• Relaxed SLOPE models can now be fit by specifying gamma in SLOPE().
• plot.trainedSLOPE() gains a new argument index, 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, @KrystynaGrzesiak!
• SLOPE() gains a new argument cd_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, and glmnet packages in
  the Suggests field that were used for testing are now removed.

SLOPE 0.5.1

Minor Changes

• Website updated to bootstrap 5-based pkgdown theme.
• Updated e-mail of maintainer.
• Dependencies on checkmate and mice were dropped.
• Update sparse matrix coercion to avoid deprecated functionality in the Matrix
  package.

SLOPE 0.5.0

Major changes

• plot.SLOPE(), plot.trainSLOPE() and plotDiagnostics() have been
  reimplemented in ggplot2.

Deprecated Functions

• caretSLOPE() has been deprecated and will be made defunct in version
  0.6.0.

SLOPE 0.4.1

Bug Fixes

• The C++ standard library memory was added to a source file to fix
  compilation errors on some systems.

SLOPE 0.4.0

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 and theta2 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 than 1e-6 to avoid
  constructions of regularization paths with infinite lambda values.
• The lambda argument in SLOPE() now also allowed the input "lasso" to
  obtain the standard lasso.
• The performance of trainSLOPE()

Vignettes

• A new vignette has been added to compare algorithms for the proximal
  operator.

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 @KrystynaGrzesiak)
• Performance of trainSLOPE() was previously hampered by erroneous
  refitting of the models, which has been fixed now (thanks @jakubkala and
  @KrystynaGrzesiak)

Deprecated and Defunct

• yvar argument in plotDiagnostics() that was previously deprecated is
  now defunct.
• Using missclass for the measure argument in trainSLOPE() has been
  deprecated in favor of misclass.

SLOPE 0.3.3

Bug fixes

• Fixed first coefficient missing from plot if no intercept was used in
  the call to SLOPE().
• Fixed incorrect results when intercept = FALSE and family = "gaussian"
  (#13, thanks, Patrick Tardivel).

SLOPE 0.3.2

Minor changes

• Added tol_rel_coef_change argument to SLOPE() as a convergence
  criterion for the FISTA solver that sets a tolerance for the relative
  change in coefficients across iterations.

Bug fixes

• Fixed premature stopping of the solver for the first step of the
  regularization path (the null model).
• Actually fix UBSAN/ASAN sanitizer warnings by modifying code for
  FISTA solver.

SLOPE 0.3.1

Bug fixes

• Fixed package build breaking on solaris because of missing STL namespace
  specifier for std::sqrt() in src/SLOPE.cpp.
• Fixed erroneous scaling of absolute tolerance in stopping criteria for
  the ADMM solvers. Thanks, @straw-boy.
• Fixed sanitizer warning from CRAN checks.

SLOPE 0.3.0

Major changes

• Scaling of alpha (previously sigma) 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 if scale = "l2" and the number of observations if
  scale = "sd" or "none". No scaling is applied when scale = "l1".
• The sigma argument is deprecated in favor of alpha in SLOPE(),
  coef.SLOPE(), and predict.SLOPE().
• The n_sigma argument is deprecated in favor of path_length in SLOPE()
• The lambda_min_ratio argument is deprecated in favor of alpha_min_ratio in
  SLOPE()
• The default for argument lambda in SLOPE() 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 argument x_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, @straw-boy.

Bug fixes

• Plotting models that are completely sparse no longer throws an error.
• rho instead of 1 is now used in the factorization part for
  the ADMM solver.

SLOPE 0.2.1

Minor changes

• A few examples in deviance() and SLOPE() that were taking
  too long to execute have been removed or modified.