nethist 1.1.0

logLik(): new S3 method for nethist, multinethist, and hnethist objects. Returns the unnormalized Bernoulli log-likelihood as a logLik object with df and nobs attributes, so AIC() and BIC() work directly on a fit. Use logLik(fit, normalized = TRUE) for the per-edge normalized value stored in fit$normalized_LL.
hnethist(): added examples to the help page.
summary(): new S3 method for nethist, multinethist, and hnethist objects. Reports the fit dimensions, the edge density, the log-likelihood with its degrees of freedom, AIC, BIC, and, for hnethist, the shape selection.

hnethist(): normalized_LL is now divided by the number of edges sum(A)/2, matching nethist() and multinethist(). Previously it was divided by sum(A), so reported values were half the size of the corresponding nethist() values. Model selection and BIC are unaffected.
print() methods for nethist, multinethist, and hnethist now respect the digits argument when printing the normalized likelihood and the mean squared error.
netsummary_plot(): the example using the suggested package network is now guarded by requireNamespace().
logLik() on a multinethist fit that collapses to a single block now reports the correct degrees of freedom.
multinethist(): method = "LSE" with multilayer input now raises an error. Previously it was accepted silently, the fit was computed by profile likelihood, and the returned object was labelled "LSE".

All fitted objects now inherit from a common base class, network_histogram: nethist fits have class c("nethist", "network_histogram"), multinethist fits c("multinethist", "network_histogram"), and hnethist fits c("hnethist", "nethist", "network_histogram"). Previously hnethist inherited from nethist while multinethist was unrelated to either.
logLik() and the print() method of the summary objects are defined once on network_histogram. summary.multinethist() and summary.hnethist() extend the base method through NextMethod(). Methods whose behaviour depends on the layer dimension, namely plot(), plot3d(), fitted(), and print(), remain specific to the subclasses.
See ?"network_histogram-class" for the documented hierarchy.
Code that tests for an exact class vector, such as identical(class(fit), c("hnethist", "nethist")), needs updating. Use inherits() instead.

The print() methods for nethist, multinethist, and hnethist are now compact. They identify the fit and report thetahat by its dimensions only, instead of printing the whole probability matrix or array, and the Available components listing has been dropped. The detailed output moved to the new summary() methods. Use object$thetahat or fitted() for the estimated probabilities.

nethist(), multinethist(), and hnethist() now state in their help pages that the greedy search is stochastic and converges to a local optimum, and that set.seed() should be called immediately before the fit for reproducible results. hnethist() additionally notes that its merging step draws stats::kmeans() starting centres from the same stream, so the selected number of shapes depends on it.
The @returns sections of nethist(), multinethist(), and hnethist() now state the full class vector including network_histogram, and multinethist() lists the method field.
Removed the leftover prompt boilerplate from the polblog and polblog_edgelist help pages, as requested by the CRAN team.
The polblog and polblog_edgelist examples now call igraph::delete_vertices() instead of the deprecated igraph::delete.vertices().

nethist 1.0.0

update: 2026/08/04

multinethist()
- New implementation of Song and Olhede (2026+) for multi-layer network histogram estimation.
- common_f option assumes a common graphon across layers.
- Accepts a wider range of inputs: 3D array, list of igraph/network objects, and ergm.multi::combined_networks object.
nethist()
- New method = "LSE" option, implementing Gao et al. (2015) for single-layer network histograms.
- Now implemented as a single-layer wrapper around multinethist().
hnethist(): new hybrid network histogram method for shape-based model selection.
fitted(): new S3 method for nethist and multinethist objects.
New dataset: IndianVil, a multilayer network.

netsummary_plot() (formerly violin_netsummary()): return 0 instead of NaN when a subsampled network is empty.

netsummary_plot(): x-axis values are now shown as igraph plots of v-shapes and cycles instead of labels.
Renamed output field LSE to MSE in nethist/multinethist objects.
Documentation terminology unified from “node” to “vertex”.
multinethist(): reject plain matrix/Matrix elements in list input, since vertex ordering across layers cannot be verified.

update: 2023/04/04

-bug fix in plot()

update: 2023/03/15

update: 2022/12/08

plot.nethist()
- You can display the estimated probabilities in p_mat of nethist objects by setting prob=TRUE. Use prob.col and prob.cex to set the color and size of text, respectively.
nethist()
- Check the convergence criterion every 5 iterations from the fifth iteration (5,10,15,…) instead of every 5 iterations from the first iteration (1,6,11,…).
- minor speed improvement.

update: 2022/11/26

Initialization step of nethist() is faster than earlier by computing hamming distance from the (simple) adjacency matrix instead of Manhattan distance.

update: 2022/11/19

nethist(): computation of the diagonal part o in greedy search Rcpp routine is fixed.
nethist(): computation of log-likelihood when there is a smaller size of group is fixed, in greedy search Rcpp routine.

update: 2022/11/08

Documentations update: provide extra about default values of arguments, rewrite examples.

update: 2022/11/04