Introduction to nethist

The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.

Overview

The nethist package estimates network histograms for single-layer and multiplex (multi-layer) networks. A network histogram partitions the nodes of a network into groups (bins) such that the connectivity pattern between groups approximates the underlying graphon — the generative probability function of the network. This provides a non-parametric, interpretable summary of large network structure.

The package implements the methods from:

Single-layer: Olhede & Wolfe (2014). Network Histograms and Universality of Blockmodel Approximation. PNAS, 111(41).
Multi-layer: Song & Olhede (2026). Joint Estimation of Sparse Multilayer Networks via Graph Limits. arXiv:2608.14536.

The package provides:

Function	Description
`nethist()`	Network histogram for a single-layer network
`multinethist()`	Network histogram for a multiplex network
`plot()`	Heatmap of the estimated histogram
`plot3d()`	3D histogram plot for multiplex networks
`summary_plot()`	Covariate distribution by estimated bin
`violin_netsummary()`	Topology summary via subgraph prevalence

Single-Layer Network Histogram

Estimating the histogram

nethist() accepts an adjacency matrix, a sparse dgCMatrix, or an igraph object. The bandwidth parameter h (number of bins) is selected automatically if not specified.

set.seed(42)
data(polblog)

# Automatic bandwidth selection
nethist_polblog <- nethist(polblog)
print(nethist_polblog)

The result is a nethist object with the following components:

cluster: a vector assigning each node to a bin (integer labels 1 to k)
thetahat: a k × k matrix of estimated connection probabilities between bins
rho_hat: estimated overall network density
normalized_LL: normalized profile log-likelihood at the solution

For a quick example with a synthetic network:

set.seed(2024)
A_gnp <- igraph::sample_gnp(200, 0.05)
result <- nethist(A_gnp)
print(result)
#> 
#> thetahat:
#>            [,1]       [,2]       [,3]       [,4]       [,5]       [,6]
#> [1,] 0.13230769 0.01331361 0.08875740 0.06213018 0.04142012 0.05621302
#> [2,] 0.01331361 0.04923077 0.09615385 0.02366864 0.04289941 0.10798817
#> [3,] 0.08875740 0.09615385 0.14461538 0.04733728 0.01775148 0.01479290
#> [4,] 0.06213018 0.02366864 0.04733728 0.09846154 0.07840237 0.00000000
#> [5,] 0.04142012 0.04289941 0.01775148 0.07840237 0.01538462 0.08431953
#> [6,] 0.05621302 0.10798817 0.01479290 0.00000000 0.08431953 0.04307692
#> [7,] 0.05856643 0.07867133 0.01398601 0.03234266 0.02010490 0.01048951
#>            [,7]
#> [1,] 0.05856643
#> [2,] 0.07867133
#> [3,] 0.01398601
#> [4,] 0.03234266
#> [5,] 0.02010490
#> [6,] 0.01048951
#> [7,] 0.07928118
#> 
#> Method: Profile Likelihood
#> 
#> normalized likelihood:
#> -3.70248405576056
#> 
#> Available components:
#> 
#> [1] "cluster"       "thetahat"      "rho_hat"       "normalized_LL"
#> [5] "MSE"           "method"        "h"

Visualising the histogram

plot() draws a heatmap of the estimated connection probabilities. Bins are shown in the order of their label by default; you can supply a custom permutation via idx_order.

plot(result)

To display the estimated probability values on each bin:

plot(result, type = "prob", prob = TRUE,
     col.regions = colorRampPalette(c("#FFFFFF", "#08306B"))(50))

Multiplex Network Histogram

A multiplex network is a collection of networks defined on the same node set, each representing a different type of relationship (layer).

Estimating the histogram

multinethist() accepts a three-dimensional adjacency array of size n × n × L, where L is the number of layers, or a list of igraph objects.

set.seed(42)
data(IndianVil)

# IndianVil is a 231 x 231 x 12 adjacency array
# representing 12 socioeconomic relationship types in an Indian village
mnethist_result <- multinethist(IndianVil)
print(mnethist_result)

The common_f argument controls whether a common histogram function is assumed across layers:

# Heterogeneous histogram (default): each layer has its own density
mnethist_het <- multinethist(IndianVil, common_f = FALSE)

# Homogeneous histogram: shared structure, layer-specific density
mnethist_hom <- multinethist(IndianVil, common_f = TRUE)

For a fast illustrative example using synthetic data:

set.seed(2024)
# Build a small 2-layer network
A1 <- igraph::as_adjacency_matrix(igraph::sample_gnp(80, 0.10), sparse = FALSE)
A2 <- igraph::as_adjacency_matrix(igraph::sample_gnp(80, 0.05), sparse = FALSE)
A_multi <- array(c(A1, A2), dim = c(80, 80, 2))

mn_result <- multinethist(A_multi)
print(mn_result)
#> 
#> Theta_hat:
#> , , 1
#> 
#>            [,1]       [,2]       [,3]       [,4]       [,5]
#> [1,] 0.15384615 0.15816327 0.03571429 0.14285714 0.14880952
#> [2,] 0.15816327 0.15384615 0.16836735 0.03571429 0.09226190
#> [3,] 0.03571429 0.16836735 0.03296703 0.12755102 0.10119048
#> [4,] 0.14285714 0.03571429 0.12755102 0.00000000 0.12202381
#> [5,] 0.14880952 0.09226190 0.10119048 0.12202381 0.01811594
#> 
#> , , 2
#> 
#>             [,1]        [,2]       [,3]        [,4]        [,5]
#> [1,] 0.021978022 0.122448980 0.08163265 0.025510204 0.005952381
#> [2,] 0.122448980 0.109890110 0.00000000 0.005102041 0.029761905
#> [3,] 0.081632653 0.000000000 0.02197802 0.056122449 0.077380952
#> [4,] 0.025510204 0.005102041 0.05612245 0.098901099 0.065476190
#> [5,] 0.005952381 0.029761905 0.07738095 0.065476190 0.050724638
#> 
#> 
#> Method: Profile Likelihood
#> 
#> normalized likelihood:
#> -3.25258572523459
#> 
#> Available components:
#> 
#> [1] "cluster"       "thetahat"      "rho_hat"       "normalized_LL"
#> [5] "MSE"           "method"        "homogeneous"   "h"

2D heatmap

plot(mn_result)

3D histogram

For multiplex networks, plot3d() displays a three-dimensional bar chart of thetahat across layers.

plot3d(mnethist_result)

Covariate Summary by Bin

Once nodes are assigned to bins, it is often informative to examine how an external covariate is distributed across bins. summary_plot() produces:

a stacked bar chart for a factor covariate
a violin plot for a numeric covariate

Factor covariate (political affiliation)

The polblog network is a hyperlink network among US political blogs. Nodes are labelled as Liberal or Conservative.

set.seed(42)
data(polblog)
nethist_polblog <- nethist(polblog)

political_label <- factor(
  c(rep("Liberal", 586), rep("Conservative", 638))
)

summary_plot(nethist_polblog, covariate = political_label,
             legend_title = "Political affiliation")

The bins with high within-group homogeneity will appear nearly solid in one colour, suggesting that the network histogram has recovered politically coherent communities.

Numeric covariate

set.seed(2024)
A_gnp <- igraph::sample_gnp(200, 0.05)
result <- nethist(A_gnp)

# Node degree as a numeric covariate
node_degree <- igraph::degree(A_gnp)
summary_plot(result, covariate = node_degree, ylab = "Degree")
#> Warning in summary_plot(result, covariate = node_degree, ylab = "Degree"): 'summary_plot' is deprecated.
#> Use 'covariate_plot' instead.
#> See help("Deprecated")

Network Topology Summary

violin_netsummary() implements the network summary statistic of Maugis et al. (2017). It estimates the prevalence of small subgraph patterns (v-shapes, triangles, squares, …) via vertex subsampling, and draws a violin plot.

This is useful for comparing the topology of an estimated histogram against the original network, or for comparing networks from different studies.

set.seed(2024)
A_gnp <- igraph::sample_gnp(400, 0.05)
violin_netsummary(A_gnp)
#> Warning in violin_netsummary(A_gnp): 'violin_netsummary' is deprecated.
#> Use 'netsummary_plot' instead.
#> See help("Deprecated")
#> Use n_rep = 574

The y-axis shows the prevalence (proportion of sampled subgraphs of that type), and the width of each violin reflects variability across subsamples. A roughly flat distribution indicates an Erdős–Rényi-like structure; peaked distributions suggest community or degree-heterogeneous structure.

Supported Input Types

All main functions accept three input formats:

Format	Class	Example
Dense adjacency matrix	`matrix`	`igraph::as_adjacency_matrix(g, sparse = FALSE)`
Sparse adjacency matrix	`dgCMatrix`	`igraph::as_adjacency_matrix(g)`
igraph object	`igraph`	`igraph::sample_gnp(200, 0.05)`

For multinethist(), the input can also be a 3D array of size n × n × L.

References

Banerjee, A., Chandrasekhar, A. G., Duflo, E., & Jackson, M. O. (2013). The diffusion of microfinance. Science, 341(6144), 1236498.

Gao, C., Lu, Y., & Zhou, H. H. (2015). Rate-optimal graphon estimation. The Annals of Statistics, 43(6), 2624–2652.

Maugis, P.-A. G., Priebe, C. E., Olhede, S. C., & Wolfe, P. J. (2017). Topology reveals universal features for network comparison. arXiv:1705.05677.

Olhede, S. C. & Wolfe, P. J. (2014). Network histograms and universality of blockmodel approximation. Proceedings of the National Academy of Sciences, 111(41), 14722–14727.

Song, Y. & Olhede, S. C. (2026). Joint Estimation of Sparse Multilayer Networks via Graph Limits. arXiv:2608.14536.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.