You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@safdari2022anomaly; @safdari2022reciprocity] have emerged as powerful tools for discovering
48
-
hidden patterns in networks, detecting communities, identifying anomalies, and generating
49
-
realistic synthetic data. However, their use is hindered by fragmented implementations, making
50
-
comparisons difficult. ProbINet addresses this critical gap by consolidating
51
-
recent approaches into a single, unified framework, allowing users to explore advanced techniques
52
-
without the overhead of navigating multiple repositories or inconsistent documentation,
53
-
boosting reproducibility and usability across disciplines.
44
+
Network analysis is central to social sciences, biology, and fraud detection, where
45
+
understanding relationships is essential. Probabilistic generative models [@contisciani2020community; @safdari2021generative; @contisciani2022community; @safdari2022anomaly; @safdari2022reciprocity] reveal hidden patterns, detect communities, identify anomalies, and generate synthetic data. Their broader use is limited by fragmented implementations that hinder comparisons and reproducibility.
46
+
ProbINet addresses this gap by unifying recent approaches in a single framework, improving accessibility and usability across disciplines.
47
+
48
+
ProbINet stands out among network analysis tools. Graph-tool [@peixoto_graph-tool_2014] provides community detection and general graph analysis tools, but it uses a different model family than our mixed-membership framework and does not account for reciprocity. CDlib [@rossetti_cdlib_2019] offers detection algorithms and evaluation routines, but ProbINet extends this with probabilistic MLE models, optional node attributes, and anomaly detection. pgmpy [@ankan_pgmpy_2024] focuses on Bayesian network structure learning, while ProbINet uncovers latent patterns like communities and reciprocity.
54
49
55
50
# Main features
56
51
57
-
ProbINet offers a versatile and feature-rich framework to perform inference on networks using probabilistic generative models. Key features include:
52
+
ProbINet offers a feature-rich framework to perform inference on networks using probabilistic
53
+
generative models. Key features include:
58
54
59
55
-**Diverse Network Models**: Integration of generative models for various network types
60
56
and goals:
@@ -83,7 +79,7 @@ ProbINet offers a versatile and feature-rich framework to perform inference on n
83
79
84
80
-**Extensible and Modular Codebase**: Future integration of additional models possible.
85
81
86
-
The **Usage** section below illustrates these features with a practical example on real-world data.
82
+
The **Usage** section below illustrates these features with a real-world example.
87
83
88
84
# Usage
89
85
@@ -94,19 +90,19 @@ directed edges representing friendships in a small Illinois high school [@konect
94
90
95
91
### Steps to Analyze the Network with ProbINet
96
92
97
-
With ProbINet, you can load network data as an edge list, select an algorithm (e.g., JointCRep),
93
+
With ProbINet, you can load network data as an edge list and select an algorithm (e.g., JointCRep),
98
94
fit the model to extract latent variables, and analyze results like soft community memberships,
99
95
which show how nodes interact across communities. This is exemplified in Figure 1. On the left, a
100
-
network representation of the input data is displayed alongside the lines of code required for its analysis using ProbINet. The resulting output is shown on the right, where nodes are colored according to their inferred soft community memberships, while edge thickness and color intensity represent the inferred probability of edge existence.
96
+
network representation of the input data is displayed alongside the lines of code required for
97
+
its analysis using ProbINet. The result is shown on the right, where nodes are colored according to their inferred soft community memberships, while edge thickness and color intensity represent the inferred probability of edge existence.
101
98
102
99
103
100
104
101
For more tutorials and use cases, see the [package documentation](https://mpi-is.github.io/probinet/).
105
102
106
103
# Running Times of Algorithms
107
104
108
-
The table below provides a general overview of the algorithms running times
109
-
on the data used in the tutorials.
105
+
The table below summarizes algorithm runtimes on the tutorial data.
110
106
**N** and **E** represent the number of nodes and edges, respectively.
111
107
Edge ranges indicate variation across layers or time steps.
112
108
**L/T** indicates the number of layers or time steps,
0 commit comments