Package 'oddnet'

Identifies anomalous networks from a series of temporal networks.

Description

This function identifies anomalous networks from a series of temporal networks. It uses graph theoretic features to transform networks to a feature space. This function has parameters for feature computation, scaling, robust PCA and anomaly detection procedures. ADD MORE DESCRIPTION.

Usage

anomalous_networks(
  networks,
  alpha = 0.05,
  dd = 2,
  trim = 0.005,
  na_action = NULL,
  vert_attr = FALSE,
  attr_name = NULL,
  attr_mat = NULL,
  fast = FALSE,
  feature_subset = NULL
)

Arguments

networks	The input series of temporal networks given in a list with each network denoted by its adjacency matrix.
alpha	An anomaly detection parameter. The level of significance for the anomaly detection algorithm lookout. Default is 0.05.
dd	A robust PCA parameter. The number of reduced dimensions in robust PCA. Default is 2.
trim	A scaling parameter. The percentage used to compute trimmed mean and trimmed standard deviation. Default is 0.5 percent.
na_action	The action for NA valued features.
vert_attr	A feature computation parameter. If TRUE the network nodes/vertices have attributes.
attr_name	A feature computation parameter. The name of the network vertex attribute. Only a single attribute can be specified.
attr_mat	A feature computation parameter. If network nodes/vertices have attributes, the list of attribute matrices for each network can be given using this feature.
fast	If set to TRUE will avoid computing time consuming features.
feature_subset	A feature computation parameter. A subset of features to be used for anomaly detection.

Value

Object imported from lookout.

See Also

[lookout::lookout()]

Examples

# We generate a series of networks and add an anomaly at 50th network.
set.seed(1)
networks <- list()
p.or.m.seq <- rep(0.1, 50)
p.or.m.seq[20] <- 0.3  # anomalous network at 20
for(i in 1:50){
  gr <- igraph::erdos.renyi.game(50, p.or.m = p.or.m.seq[i])
  networks[[i]] <- igraph::as_adjacency_matrix(gr)
}
anomalous_networks(networks, fast = TRUE)

---

Computes features for each network.

Description

This function computes features for each network using graph theoretic constructs.

Usage

compute_features(gr, attributes = FALSE, attr_name = NULL, fast = FALSE)

Arguments

gr	The network or graph as an igraph object.
attributes	If the network nodes/vertices have attributes, then attributes = TRUE.
attr_name	The name of the node/vertex attribute. Only a single attribute can be specified.
fast	If set to TRUE will avoid computing time consuming features.

Value

A network features object containing 20 graph-theoretic features.

Examples

set.seed(1)
gr <- igraph::erdos.renyi.game(100, 0.05)
compute_features(gr)

---

Laplacian Eigen Value method by Shenyang Huang, Yasmeen Hitti, Guillaume Rabusseau and Reihaneh Rabbany from their KDD'20 paper Laplacian Change Point Detection for Dynamic Graphs

Description

Laplacian Eigen Value method by Shenyang Huang, Yasmeen Hitti, Guillaume Rabusseau and Reihaneh Rabbany from their KDD'20 paper Laplacian Change Point Detection for Dynamic Graphs

Usage

lad(matlist, k = NULL, short_win, long_win, alpha = 0.05, from_file = NULL)

Arguments

matlist	The matrix list, where each matrix is an adjacency matrix of the graph.
k	The number of eigen values to connsider
short_win	The length of the shorter windows
long_win	The length of the longer windows
alpha	The threshold to declare anomalies
from_file	This is an additional parameter only if a file needs to be read

Value

An object of class lad. LAD is a window based method. It considers short and a long windows. The lad object has anomalous scores when taking into account short and long windows along with the identified anomalies for both short and long windows.

References

Huang, S., Hitti, Y., Rabusseau, G., & Rabbany, R. (2020). Laplacian Change Point Detection for Dynamic Graphs. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 349–358. https://doi.org/10.1145/3394486.3403077

Examples

# We generate a series of networks and add an anomaly at 50th network.
set.seed(1)
networks <- list()
p.or.m.seq <- rep(0.05, 50)
p.or.m.seq[20] <- 0.2  # anomalous network at 20
for(i in 1:50){
  gr <- igraph::erdos.renyi.game(100, p.or.m = p.or.m.seq[i])
  networks[[i]] <- igraph::as_adjacency_matrix(gr)
}
ladobj <- lad(networks, k = 6, short_win = 2, long_win = 4)
ladobj

---