Changelog (original) (raw)

Documentation:

• Improved documentation and added new pkgdown documentation website.
• NEWS file moved to markdown format instead of .Rd and available on the website

Adjustments:

• Documentation roxygenized.
• More selective importing and exporting.
• S3 implementation of ridgeP output.

Documentation:

• Updated CITATION file
• Updated README file

Bug fixes:

• Fixed bug in GGMpathStats: Incorrectly stated before that all igraph layouts were supported. Now they indeed are supported.

Adjustments:

• Bioconductor dependencies are now automatically installed upon first installation of rags2ridges.
• GGMpathStats now has additional visualization options: It can handle all layout functions supported by igraph. Moreover, it is now possible to specify custom coordinates for node-placement.

Notifications:

• Hot fix due to class changes in “matrix”. No major visible user changes.
• CNplot function again updated: higher max. iterations for Lanczos method

Notifications:

• Hot fix due to new RNG. No visible user changes.

Notifications:

• optPenalty.LOOCV is deprecated. Please use optPenalty.kCV instead
• optPenalty.LOOCVauto is deprecated. Please use optPenalty.kCVauto instead

Documentation:

• Updated CITATION file
• Updated README file

Adjustments:

• sparsify now has an additional thresholding option: ‘connected’

Documentation:

• Updated CITATION file
• Updated README file

Bug fixes:

• Fixed bug in Ugraph:
  • Incorrectly stated before that all igraph layouts were supported.
  • Now they indeed are supported.

Notifications:

• conditionNumberPlot is deprecated. Please use CNplot instead
• Features of the CNplot function (above and beyond conditionNumberPlot):
  • The digitLoss and rlDist arguments have been removed
  • These arguments have been replaced with the logical argument Iaids
  • Iaids = TRUE amends the basic condition number plot with interpretational aids
  • These aids are the approximate loss in digits of accuracy and and approximation of the acceleration along the regularization path of the condition number
  • Argument main is now a character argument
  • Argument value now by default takes the value 1e-100 (convenient)
  • Now uses C++ functionalty for additional speed

Adjustments:

• edgeHeat now has aligned x-axis labels
• The visualizations of the optPenalty.LOOCV and optPenalty.aLOOCV functions now will no longer produce horizontal and/or vertical lines that fall outside the boundaries of the figure
• optPenalty.LOOCV now uses log-equidistant penalty grid for optimal penalty parameter determination (this also enhances the visualization)
• New features updated optPenalty.aLOOCV function:
  • Function has been sped up by killing redundant inversion
  • now uses log-equidistant penalty grid for optimal penalty parameter determination (this also enhances the visualization)
• New features updated Ugraph function:
  • One can now also specify vertex placement by coordinate specification
  • Now outputs, for convenience, the vertex coordinates of the plotted graph
• ridgePathS has been sped up by killing redundant inversion
• The covML function has been amended with an argument that indicates if a correlation matrix (instead of an ML estimate of a covariance matrix) is desired. This offers more flexibility. One can now get the ML estimate of the covariance matrix, the ML estimate of the covariance matrix on standardized data, as well as the correlation matrix
• The optPenalty.LOOCVauto function has been amended with an argument that indicates if the evaluation of the LOOCV score should be performed on the correlation scale
• The optPenalty.LOOCV function has been amended with an argument that indicates if the evaluation of the LOOCV score should be performed on the correlation scale
• The optPenalty.aLOOCV function has been amended with an argument that indicates if the evaluation of the approximate LOOCV score should be performed on the correlation scale

Documentation:

• Added this NEWS file!
• Updated (and corrected) CITATION file
• Added README file
• Added (selective) import statements for default packages as required for R-devel

Additions:

• rags2ridges now uses Rcpp and RcppArmadillo with core functions written in C++. The package should now be at least two orders of magnitude faster in most cases.
• Added, next to the core module, the fused ridge module. The fused module provides functionality for the estimation and graphical modeling of multiple precision matrices from multiple high-dimensional data classes. Functions from this module are generally suffixed with .fused. Functions tied to (or added with) this module are:
  • isSymmetricPD
  • isSymmetricPSD
  • is.Xlist
  • default.target.fused
  • createS
  • getKEGGPathway
  • kegg.target
  • pooledS
  • pooledP
  • KLdiv.fused
  • ridgeP.fused
  • optPenalty.fused.grid
  • print.optPenaltyFusedGrid
  • plot.optPenaltyFusedGrid
  • optPenalty.fused.auto
  • optPenalty.fused
  • default.penalty
  • fused.test
  • print.ptest
  • summary.ptest
  • hist.ptest
  • plot.ptest
  • sparsify.fused
  • GGMnetworkStats.fused
  • GGMpathStats.fused
• The following functions were added to the core module:
  • covMLknown
  • GGMmutualInfo
• Added miscellaneous (hidden) functions.

Bug fixes:

• Fixed bugs in GGMpathstats:
  • Code no longer breaks down if variable names are absent.
  • Now properly handles singleton pathsets.
• Fixed bug in sparsify: Now always returns symmetric objects

Adjustments:

• Argument verticle as used in various functions has been renamed to vertical. Sorry for any inconvenience.
• Internal usage of ridgeS replaced by the faster C++-dependent counterpart ridgeP
• New features updated conditionNumberPlot function:
  • Function has been sped up
  • Now uses log-equidistant grid for visualization
  • Now includes option to additionally plot the approximate loss in digits of accuracy

Notifications:

• ridgeS is deprecated. Please use ridgeP instead
• Future versions of rags2ridges will be subject to changes in naming conventions

Additions:

• Inclusion hidden function .pathContribution for usage in GGMpathStats function
• Inclusion hidden function .path2string for usage in GGMpathStats function
• Inclusion hidden function .pathAndStats for usage in GGMpathStats function
• Inclusion hidden function .cvl for usage in optPenalty.LOOCVauto function
• Inclusion hidden function .lambdaNullDist for usage in GGMblockNullPenalty function
• Inclusion hidden function .blockTestStat for usage in GGMblockTest function
• Inclusion function that expresses the covariance between a pair of variables as a sum of path weights: GGMpathStats
• Inclusion function that determines the optimal penalty parameter value by application of the Brent algorithm to the LOOCV log-likelihood: optPenalty.LOOCVauto
• Inclusion function that generates the distribution of the penalty parameter under the null hypothesis of block independence: GGMblockNullPenalty
• Inclusion function that performs a permutation test for block structure in the precision matrix: GGMblockTest
• Inclusion wrapper function: fullMontyS

Bug fixes:

• Corrected small error in evaluateSfit function. The dir argument was not properly used previously.

Adjustments:

• New features updated optPenalty.aLOOCV function:
  • For scalar matrix targets the complete solution path depends on only 1 eigendecomposition and 1 matrix inversion. Meaning: the function is sped up somewhat by lifting redundant inversions out of for loops.
  • Optional graph now plots the approximated LOOCV negative log-likelihood instead of ln(approximated LOOCV negative log-likelihood).
  • Legend in optional graph has been adapated accordingly.
• New features updated optPenalty.LOOCV function:
  • Optional graph now plots the LOOCV negative log-likelihood instead of ln(LOOCV negative log-likelihood).
  • Legend in optional graph has been adapated accordingly.
• New features updated default.target function:
  • Inclusion new default target option: type = DIAES. Gives diagonal matrix with inverse of average of eigenvalues of S as entries.
• New features updated GGMnetworkStats function:
  • Now also assesses (and returns a logical) if graph/network is chordal.
  • Now also includes assesment of the eigenvalue centrality.
  • Now includes option to have list or table output.
• New features updated ridgePathS function:
  • Sped up considerably for rotation equivariant alternative estimator. By avoidance of redundant eigendecompositions and inversions.
  • Now catches breakdown due to rounding preculiarities when plotType = "pcor".
• New features updated sparsify function:
  • Inclusion new thresholding function top: retainment of top elements based on absolute partial correlation.
  • Inclusion output option: When output = "light", only the (matrix) positions of the zero and non-zero elements are returned.
  • Function no longer dependent on GeneNet; now makes direct use of fdrtool.
  • Function now also prints some general information on the number of edges retained.

Additions:

• Inclusion hidden function .ridgeSi for usage in conditionNumberPlot function.
• Inclusion hidden function .eigShrink for usage in (a.o.) ridgeS function.
• Inclusion function calculating various network statistics from a sparse matrix: GGMnetworkStats
• Inclusion function for visual inspection fit of regularized precision matrix to sample covariance matrix: evaluateSfit
• Inclusion function for visualization of regularization paths: ridgePathS
• Inclusion function for default target matrix generation: default.target

Adjustments and name changes:

• New features updated evaluateS function:
  • The printed output of the evaluateS function is now aligned
  • Calculation spectral condition number has been improved
• conditionNumber function now called conditionNumberPlot. The updated function has new features:
  • Main plot can now be obtained with either the spectral (l2) or the (approximation to) l1 condition number
  • Main plot can now be amended with plot of the relative distance to the set of singular matrices
  • The title of the main plot can now be suppressed
  • One can now obtain numeric output from the function: lambdas and condition numbers
• New features updated sparsify function:
  • Changed type = c("threshold", "localFDR") to threshold = c("absValue", "localFDR") (clarifying nomenclature)
  • Changed threshold to absValueCut (clarifying nomenclature)
  • Will now output both sparsified partial correlation/standardized precision matrix and the sparsified precison matrix, when input consists of the unstandardized precision matrix
• New features updated ridgeS function:
  • Contains an improved evaluation of the target matrix possibly being a null matrix
  • Now evaluates if a rotation equivariant alternative estimator ensues for a given target matrix
  • When rotation equivariant alternative estimator ensues, computation is sped up considerably by circumventing the matrix square root
• optPenaltyCV function now called optPenalty.LOOCV, for sake of (naming) consistency. The updated function has new features:
  • targetScale option has been removed
  • Replaced log in optional graph by ln
  • Visual layout of optional graph now more in line with recommendations by Tufte (regarding data-ink ratio)
• New features updated optPenalty.aLOOCV function:
  • Replaced log in optional graph by ln
  • Visual layout of optional graph now more in line with recommendations by Tufte (regarding data-ink ratio)
• Computation optimal penalty in conditionNumberPlot, optPenalty.aLOOCV and optPenalty.LOOCV functions sped up considerably for rotation equivariant alternative estimator. By usage new ridgeS and avoidance of redundant eigendecompositions
• Default target in ridgeS, conditionNumberPlot, optPenalty.aLOOCV and optPenalty.LOOCV now \code{“DAIE” option from default.target

Additions:

• Inclusion function for ML estimation of the sample covariance matrix: covML
• Inclusion function for approximate leave-one-out cross-validation: optPenalty.aLOOCV
• Inclusion function conditionNumber to visualize the spectral condition number over the regularization path
• Inclusion function evaluateS to evaluate basic properties of a covariance matrix
• Inclusion function KLdiv that calculates the Kullback-Leibler divergence between two normal distributions
• Inclusion option to suppress on-screen output in sparsify function

Bug fixes:

• Corrected small error in optPenaltyCV function

Adjustments:

• Both optPenaltyCV and optPenalty.aLOOCV now utilize covML instead of cov
• Default output option in optPenaltyCV (as in optPenalty.aLOOCV) is now light