Type: Package
Title: Graph Prediction from a Graph Time Series
Version: 0.1.2
Maintainer: Sevvandi Kandanaarachchi <sevvandik@gmail.com>
Description: Predicting the structure of a graph including new nodes and edges using a time series of graphs. Flux balance analysis, a linear and integer programming technique used in biochemistry is used with time series prediction methods to predict the graph structure at a future time point Kandanaarachchi (2025) <doi:10.48550/arXiv.2507.05806>.
License: GPL (≥ 3)
Encoding: UTF-8
RoxygenNote: 7.3.2
Imports: dplyr, fable, fabletools, forecast, future, igraph, lpSolve, Matrix, rlang, stats, tibble, tsibble
Suggests: feasts, knitr, nnet, rmarkdown, urca
URL: https://sevvandi.github.io/netseer/
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2025-09-09 23:20:40 UTC; kan092
Author: Sevvandi Kandanaarachchi ORCID iD [aut, cre], Stefan Westerlund [aut]
Repository: CRAN
Date/Publication: 2025-09-10 07:50:32 UTC

Generates a bigger graph using exponential growth.

Description

Generates a bigger graph using parameters for node and edge growth. If a sequence of graphs are created, the number of nodes in this sequence would exponentially increase.

Usage

generate_graph_exp(
  gr = NULL,
  del_edge = 0.1,
  new_nodes = 0.1,
  edge_increase = 0.1
)

Arguments

gr

The input graph to generate the next graph. If set to NULL a graph using igraph::sample_pa is used as the input graph.

del_edge

The proportion of edges deleted from the input graph. Default set to 0.1.

new_nodes

The proportion of nodes added to the input graph. Default set to 0.1.

edge_increase

The proportion of edges added to the input graph. Default set to 0.1.

Value

A graph.

Examples

set.seed(1)
gr <- generate_graph_exp()
gr

Generates a bigger graph by linear growth.

Description

Generates a bigger graph using parameters for node and edge growth. If a sequence of graphs are created, the number of nodes would linearly increase.

Usage

generate_graph_linear(
  gr = NULL,
  del_edge = 1,
  new_nodes = 1,
  edge_increase = 1,
  edges_per_new_node = 3
)

Arguments

gr

The input graph to generate the next graph. If set to NULL a graph using igraph::sample_pa is used as the input graph.

del_edge

The number of edges deleted from the input graph. Default set to 1.

new_nodes

The number of nodes added to the input graph. Default set to 1.

edge_increase

The number of edges added to the input graph. Default set to 1.

edges_per_new_node

The number of edges added to the new nodes. Default set to 3.

Value

A graph.

Examples

set.seed(1)
gr <- generate_graph_linear()
gr

Predicts a graph from a time series of graphs.

Description

This function predicts the graph at a future time step using a time series of graphs.

Usage

predict_graph(
  graphlist,
  formulation = 2,
  conf_level1 = NULL,
  conf_level2 = 90,
  dense_opt = 2,
  weights_opt = 8,
  weights_param = 0.001,
  h = 1
)

Arguments

graphlist

A list of graphs in igraph format.

formulation

Formulation 2 includes an additional condition constraining total edges by the predicted value. Formulation 1 does not have that constraint. Formulation 2 gives more realistic graphs due to that constraint. Default is set to 2.

conf_level1

A value between 50 and 100 denoting the confidence interval for the number of predicted nodes in the graph. If set to NULL the predicted graph has the mean number of predicted nodes. If set to 80 for example, there would be 3 predicted graphs. One with mean number of predicted nodes, and the other two with the number of nodes corresponding to lower and upper confidence bounds.

conf_level2

The upper confidence bound for the degree distribution. Default set to 90.

dense_opt

If set to 2 the dense option in R package lpSolve will be used.

weights_opt

Weights option ranging from 1 to 6 used for different edge weight schemes. Weights option 1 uses uniform weights for all edges. Option 2 uses binary weights. If the edge existed in a past graph, then weight is set to 1. Else set to 0. All possible new edges are assigned weight 1. Option 3 is a more selective version. Option 4 uses proportional weights according to the history. Option 5 uses proportional weights, but as the network is more in the past, it gives less weight. Option 5 uses linearly decaying proportional weights. Option 6 uses harmonically decaying weights. That is the network at T is given weight 1, T-1 is given weight 1/2 and so on. Option 7 uses 1 for edges that are present in the last graph. Option 8 is a slightly different to Option 7. It uses 1 for edges in the last seen graph and the weights_param for new edges. Default is set to 8.

weights_param

The weight given for possible edges from new vertices. Default set to 0.001.

h

The prediction time step. Default is h = 1.

Value

A list of predicted graphs. If conf_level1 is not NULL, then 3 graphs are returned one with the mean number of predicted nodes and the other 2 with the number of nodes equal to the lower and upper bound values of prediction. If If conf_level1 is NULL, only the mean predicted graph is returned.

Examples

set.seed(2024)
edge_increase_val <- new_nodes_val <- del_edge_val <- 0.1
graphlist <- list()
graphlist[[1]] <- gr <-  igraph::sample_pa(5, directed = FALSE)
for(i in 2:15){
  gr <-  generate_graph_exp(gr,
                        del_edge = del_edge_val,
                        new_nodes = new_nodes_val,
                        edge_increase = edge_increase_val )
  graphlist[[i]] <- gr
}
grpred <- predict_graph(graphlist[1:15], conf_level2 = 90, weights_opt = 6)
grpred