NEWS

backbone 3.0.2

remove deprecated functions from manual
replace all remaining %*% with (t)crossprod()
include backbone details as graph attributes when input is an igraph object
add backbone() wrapper function for all input network types
speed-ups to .fixedfill() null model
specifying backbone_only = FALSE returns a backbone-class object that can be described using print()

all functions re-written to be modular, to facilitate future extensions
streamlined functions to focus on input network type, rather than backbone model
keep attributes of retained edges in igraph objects
functions renamed in snake_case, to match naming conventions in igraph
eliminated support for edgelist inputs, because they can be ambiguous
eliminated ordinal stochastic degree sequence model (oSDSM) for bipartite projections, because it has limited applications and has not been formally validated
all functions have associated unit tests

speedups in sparsify() and all statistical backbone functions
eliminated hyperg() as alternate name for fixedrow(), eliminated universal() as alternate name for global()
empty & full rows/cols no longer need to be removed from bipartite inputs
replaced testthat with tinytest; expanded unit tests
backbone object includes node attributes, if present

eliminated dependency on PoissonBinomial; sdsm() and fixedcol() now use an efficient implementation of the Refined Normal Approximation in base R
eliminated dependency on MASS; osdsm() now uses glm() in base R to implement the conditional logistic regression method described by Neal (2017)
eliminated dependency on network and support for network objects, which can easily be converted to matrix objects
removed bipartite generative functions bipartite.from.probability(), bipartite.from.sequence(), bipartite.from.distribution(), and bipartite.add.blocks(). These are now part of the incidentally package
speed improvements to bicm()
updated the information provided in the narrative text when narrative = TRUE
when the original graph is supplied as an igraph object with vertex attributes, the attributes are preserved in the backbone
added links to new tutorial: Neal, Z. P. 2022. backbone: An R Package to Extract Network Backbones. PLOS ONE, 17, e0269137. https://doi.org/10.1371/journal.pone.0269137

minor bug fixes
faster implementation of fastball() algorithm
set alpha = 0.05 as default in all statistical models
renamed fwer (familywise error rate) parameter as mtc (multiple test correction)

remove davis example data; add examples using synthetic data
add support for unweighted graphs: sparsify()
add support for weighted bipartite graphs: osdsm()
add support for non-projection weighted graphs: disparity()
new vignette illustrating all functions
add implementation of fastball() algorithm for marginal-preserving matrix randomization
re-add testthat tests
allow backbone functions to directly output a backbone, eliminating the need for the backbone.extract() function
add support for any p.adjust() method of correcting for familywise error rates
Minor bug fixes

removed testthat tests due to unknown MKL error; will be restored in a future version

add four functions to generate random bipartite graphs: bipartite.from.probability(), bipartite.from.sequence(), bipartite.from.distribution(), and bipartite.add.blocks()
set diagonal in positive and negative backbone object matrices to NA
corrected p-value computation in fixedfill()
remove running time from backbone object summary dataframe
update documentation, readme, vignette

add fixedcol() function - null model where column degrees are fixed and row sums are allowed to vary
add fixedfill() function - null model where the number of 1’s in the matrix (number of edges in the graph) are fixed
replace class.convert() with tomatrix() and frommatrix()
use updated Poisson binomial calculations (more accurate approximation)
hyperg() now called fixedrow()
remove bipartite.null function
update documentation, readme, vignette
include logo

update sdsm to use the bicm model - a new, fast, approximation of the probabilities
remove all other models from sdsm
if an older model is called in sdsm, show warning that model has changed
add new function bipartite.null which lets the user pick if they want rows/cols to be fixed or vary
update fwer m parameter

fix fdsm to accept all graph inputs
rename sdsm “chi2” model to “rcn”
universal function can now return weighted projection
universal function now has a narrative parameter
class.convert now drops (with warning) rows and columns with zero sum before sending output to universal, sdsm, fdsm, or hyperg.
update citations

add support for sparse matrix, igraph, network, and edgelist objects (see ‘class.convert’)
add family-wise error rate test corrections (see ‘backbone.extract’)
sdsm: add multiple methods for computing initial probabilities (see ‘sdsm’ details) one of which uses convex optimization (see ‘polytope’)
sdsm: update poisson binomial computation method to increase speed (see ‘sdsm’ and ‘rna’)
add more descriptives to summary dataframe output of backbone object
update documentation of functions
update vignette to reflect package changes
bug fixes for R 4.0.0