Empirical statistical analysis, visualization and simulation of
diffusion and contagion processes on networks. The package implements algorithms
for calculating network diffusion statistics such as transmission rate, hazard
rates, exposure models, network threshold levels, infectiousness (contagion),
and susceptibility. The package is inspired by work published in Valente,
et al., (2015)
CHANGES IN NETDIFFUSER VERSION 1.20.0 (2018-06-06)
New functions and features
diffreg provides a wrapper of glm to run lagged regression models.
Default colors for plot_diffnet2, plot_infectsucept, and others are now selected from the viridis R package, which provides perceptually uniform and colorblind proof colors.
plot_diffnet2 now has the correct scaling in nodes positions.
rdiffnet_multiple calls library(netdiffuseR) when using multicore.
arrow.width in plot_threshold.default now set to be equal to nslices(graph)/80.
curved option passed to plot_._threshold.default.
The c++ function edges_arrow now returns two different elements (the edge, and the arrow).
CHANGES IN NETDIFFUSER VERSION 1.19.0 (2017-10-16)
New functions and features
rdiffnet now allows passing scalars for threshold.dist, more over, the user can also ask the function to just warn when there is no diffusion instead of returning with error.
plot.diffnet, plot_diffnet, plot.diffnet_mentor, and plot_diffnet2 use igraph for plotting. Also, users can now pass "degree" for -vertex.size- (vertex.cex not used anymore), allowing to "automatically" scale vertices by (in/out/.)degree. Also, plotting arguments like vertex.* or edge.* are standarized so these match igraph.
plot_diffnet has a new parameter: background.
rdiffnet_multiple, a wrapper of rdiffnet, allows performing simulations studies by running simulating multiple diffusion networks using rdiffnet.
exposure has a new parameter: lags. By default lags = 0, returns a lagged exposure matrix.
igraph_to_diffnet was failing with the graph had no weights.
drop_isolated was not behaving well for diffnet objects.
vertex_covariate_dist was incorrectly specified. Only the default p=2 were OK. Now fixed and the tests/ folder includes a tests on this.
plot_diffnet2 was not passing color.ramp to drawColorKey. Now fixed.
plot.diffnet_mentor had a bug. Uncesary permutation of vertices was done, but it actually had no visible effect. Similar problem was corrected in diffnet_to_igraph, and other plot methods using igraph for plotting.
Replacing some C++ functions by R functions in cases in which there was no decrease in performance.
plot_diffnet function now has smaller margins, so looks more appealing.
New examples in vignettes "netdiffuseR showcase: Medical Innovations", and "Simulating diffusion networks: Using the rdiffnet function".
CHANGES IN NETDIFFUSER VERSION 1.18.1 (2017-07-22)
network_to_diffnet, diffnet_to_network coercion between diffnet and network objects.
networkDynamic_to_diffnet, diffnet_to_networkDynamc coercion between diffnet and networkDynamic objects.
diffnet_to_igraph was copying over a single adjacency matrix, which did a difference in dynamic networks.
diffnet_to_igraph was not considering loops correctly.
In unexpected situations egonet_attrs was crashing.
CHANGES IN NETDIFFUSER VERSION 1.18.0 (2017-07-16)
bootnet implements network bootstrapping based on Snijders and Borgatti (1999)
mentor_matching implements Valente and Davis (1999) Mentor matching algorithm. including a plot method.
approx_geodesic an alternative to igraph::distances and sna::geodist. computes geodesics up to a certain number of steps and returns a sparse matrix.
matrix_compare Efficiently compares two sparse matrices looking only at the valued cells.
as_dgCMatrix Coerce matrix-like objects into dgCMatrix objects (sparse matrices from the Matrix package).
fitbass Fits the Bass Diffusion Model to an observed vector of cumulative adopters. The estimation is done via stats::nls.
netmatch and netmatch_prepare (on development) implement matching estimators with network data.
dgr returned with error when self == TRUE
In some calls to igraph::graph_from_adj... sorting of vertices was not preserved.
The matrix method in egonet_attrs was passing a list of vertices instead of the attributes. Fixed.
transformGraphBy was returning with error when the time periods ranged other than 1, 2, ...
rgraph_er is now significantly faster (orders of magnitude compare to previous versions). rgraph_ba is faster too.
moran now returns the sd, expected and p-value.
exposure now receives static graphs in alt.graph with a warning.
rewire_graph now also uses QAP. This affects directly to structu_test.
CHANGES IN NETDIFFUSER VERSION 1.17.0 (2016-11-10)
New features and changes
The title of the package is now Analysis of Diffusion and Contagion Processes on Networks.
The function struct_test now allows other types of graphs. Before it only supported diffnet objects.
The function rewire_graph gains a new argument for the algorithm "swap". Now to ensure aperiodicity in MCMC a chance of skiping a rewire has been included.
The function n_rewires now has a default of 20 (before it was 100). This is based on Ray et al (2012) (more details in the manual).
The function rgraph_ba gains a new argument, self=TRUE. By default behaves as before following Bollobas, but now can deviate to generate graphs with no autolinks.
In rgraph_ba, the argument eta allows implementing De Almeida et al. (2013) Scale-free homophilic networks.
The functions exposure and dgr are now pure R code (C++ functions were replace since there were no significant speed gains).
diffnet class objects now have two new meta-values: name and behavior.
Elements -graph-, -toa-, -adopt- and -cumadopt- in diffnet class objects have lost their dimnames (more efficient storage).
classify_adopters now always includes Non-Adopters.
vertex_covariate_dist computes distances between vertices using both the graph and a matrix of length nxK.
vertex_mahalanobis_dist computes mahalanobis distance between vertices (as above).
struct_test_asymp an asymptotic approximation of struct_test (not recomended).
ego_variance computes a pseudo variance at the ego level (aux function for struct_test).
transformGraphBy applies a function that transforms a graph considering structural zeros given by groups. Similar to the idea of the -by- option in struct_equiv.
read_ucinet read UCINET binary files (both header and graph file). Still work in progress.
plot.diffnet_degSeq method allows visualizing degree sequence as log-log plots (default).
diag_expand creates a single adjacency matrix from a dynamic graph.
summary.diffnet_adoptChange method generates a summary table of the df generated by select_egoalter.
permute_graph permutes the values of an adjacency matrix (Conditional Uniform Graph).
rewire_qap generates isomorphic graphs by "changing the labels".
^.diffnet method was rasing to +1 power, e.g. diffnet^2 was actually diffnet^3.
/.diffnet was not working.
plot_diffnet was computing the coordinates of the cells wrongly. Most of the time causing adding figures outside of the plotting area.
The c.diffnet_struct_test method was not updating the p.value.
The function edgelist_to_adjmat was not processing correctly undirected graphs when the edgelist represented a lower triangular matrix.
The function survey_to_diffnet had an issue processin dyn graph attrs returning errors. Now fixed.
The function select_egoalter returned error when graph was an array.
The method [[<-.diffnet failed when replacing a dynamic attribute with a NULL value (e.g. dn[["my_dyn_att"]] <- NULL).
CHANGES IN NETDIFFUSER VERSION 1.16.7 (2016-07-07)
Fixed bug in struct_equiv: When groupvar was a list (dynamic attr), the function returned error.
Fixed bug in rewire_graph.array: Returned error when algorithm="swap"
Fixed bug in rewire_graph: The option copy.first was not been applied correctly.
In hist.diffnet_struct_test: ... now passed to hist.default.
Fixed bug on egonet_attrs: The matrix method was returning with error.
New features and changes
plot_infectsuscept includes 2D kernel smoother via MASS::kde2d.
infection, susceptibility and threshold now report NA for non-adopters or excluded variables.
egonets_attrs now has new argument: self.attrs allows including ego's attributes as part of the outcome so it can be used by the user.
plot_diffnet now uses igraph::plot.igraph for plotting instead
threshold gains a new argument: lags now users can define threshold as exposure lags time periods prior to the time of adoption. By default is 0 so its exposure at the time of adoption.
New method c.diffnet_struct_test: A wrapper of boot:::c.boot.
diffusionMap computes the required matrix to be used with image-like functions mapping a vertex covariate given a graph structure.
n_rewires: computes a suggested number of rewires per step in order to attain unbiased graph samples.
diffnetLapply: Apply throught periods on diffnet objects.
Several new methods for the class diffnet. Now users can apply str, dimnames (so colnames and rownames), t, &, |, dim and %*%.
drawColorKey: Handy function to draw a color key in the current plot.
classify_adopters: As in Valente (1995), depending on time of adoption, adopters are classified as early adopters, early majority, late majority, and laggards. The function introduces a new class and has methods for ftable and plot.
rescale_vertex_igraph: Helper function to fix the size of vertex when calling plot.igraph so that the size is proportional to the x-axis.
CHANGES IN NETDIFFUSER VERSION 1.16.5 (2016-05-02)
Bug fixed on edgelist_to_adjmat: Counting number of vertices is now done right after recode. (Reported by Tom)
Fixed bug in diffnet.attrs(..., as.df=TRUE). ids were wrongly retrieved.
Fixed bugs for rgraph_ba_cpp: Degree of new vertices was not changing apropiately. This only was an issue when m>1.
Fixed bugs for the as_diffnet method for arrays.
Fixed bugs in rewire_graph. Indexing of the jth component (when rewiring) was not been made correctly (now it does). Also, when rewiring, the new endpoints were truncated to n-1 (now fixed).
Fixed bugs for as_diffnet: When a dynamic graph was passed with slices names different from the time periods, the slices names were kept. Now these are replaced by meta$pers.
New features and changes
Support for int64_t in RcppArmadillo now allows for creating/reading adjacency matrices with more than 4 billion elements (big graphs).
In edgelist_to_adjmat use.incomplete has been replaced by keep.isolated which makes more sense for naming. Incomplete cases on times or weights are still ignored (as these cannot be processed by the c++ 'engine'). (Reported by Tom)
In edgelist_to_adjmat times has been replaced by t0 and t1. So now the user can import graphs with spells.
Added new elements to the diffnet_struct_test class: p.value, t0, mean_t, and R. All these were available before either to be computed or retrieved from the boot list at the class.
New argument for threshold. Now, by default, threshold levels are not computed for adopters in the first time period as this can be a biased estimate. If the user wants to compute such, he/she can set include_censored=TRUE.
Attributes in diffnet objects are now stored as data frames (instead of matrices). This affects the function diffnet.attrs, and egonet_attrs as these use attributes directly. (Requested by Tom)
New features for the rewire_graph function. In particular, p can now be a vector of length T, so each slice can have different rewiring prob., and the new option copy.first which allows to recycle the first rewired slice (see details).
New features for the exposure function. When graph is of class diffnet, the function accepts attrs equal to the name of some the graph's attributes. Also, alt.graph can be specified as se, which will be replaced by the inverse of the structural equivalence. When valued=FALSE the function will switch it to TRUE and warn the user.
New argument for struct_equiv and exposure, groupvar. This new option provides a convenient way of calculating structural equivalence and exposure clusterized by group. Specially useful when there are different communities in a graph. See examples in the manual.
as_diffnet now has an internal function, check_as_diffnet_attrs, to check input attributes dimensions and coerce them into proper class/structure. Valid attributes are now documented in the function's manual.
New arguments for edges_coords: dev and ran allow including device + margins aspec ratio and plotting area y/x limits for improved aspect ratio computation.
New internal function edges_arrow: Computes the coordinates of a 4 points polygon allowing to draw pretty arrows considering aspec ratio of device, margins and y/x.
Geodesic distances are now computed using igraph::distances instead of sna::geodist as it is more flexible and faster.
New arguments for plot_threshold: vertex.sides, vertex.lab.cex, vertex.lab.adj, vertex.lab.col, vertex.rot, jitter.factor, and jitter.amount to give more control.
New internal function vertex_coords: Creates polygons of any given number of sides considering aspec ratio of both x/y and device.
New features for rdiffnet. seed.graph can be either a function that generates a random graph, a character string (as before) indicating the class of graph to generate, or any other class of graph (either static or dynamic) as specified in netdiffuseR-graphs. seed.nodes can now be a vector with indices pointing to the initial adopters.
The rewiring algorithm for rgraph_ws has been replaced with a rewire_ws which has been implemented as it was presented in Watts and Strogatz (1998).
New function: survey_to_diffnet. This function allows importing network nomination data (in survey fashion) of both types, cross-section and panel formats (static network only varying adoption, or dynamic network varying attributes and network structure simultaneously).
New function: edgelist_to_diffnet. Similar to survey_to_diffnet, this function reads diffusion networks from an adjacency matrix and a vertex attributes data frame. Both the attributes and the edgelist can be static or dynamic.
New method: as.array.diffnet.
New functions: read_pajek and read_ucinet. Still on development.
New functions: nvertices and nedges return the number of vertices and edges that a graph has. This can be applied to any class of graph accepted by the package.
New indexing methods via [[.diffnet, [[<-.diffnet, for network attributes and [.diffnet and [<-.diffnet for adjacency matrix. The function diffnet.attrs<- will be deprecated for the next CRAN release. The function diffnet.subset.slices is now not exported (internal use), so the user needs to use the [.diffnet method instead.
New concatenating method c.diffnet for diffnet objects. This method allows 'adding up' diffnet objects.
New print method for diffnet_se, objects returned by struct_equiv.
New function diffnet_to_igraph.
New rewiring algorithm, rewire_swap implements the edge-switch algorithm in an efficient way. This preserves degree sequences.
CHANGES IN NETDIFFUSER VERSION 1.16.2 (2016-02-18)