tendai-zw / anomalous-vertices-detection Goto Github PK

This project forked from kagandi/anomalous-vertices-detection

Novel anomalous vertices detection method.

Makefile 0.08% Python 99.92%

anomalous-vertices-detection's Introduction

The Anomalous-Vertices-Detection project is a Python package for performing graph analysis. The package supports extracting graphs' topological features, performing link prediction, and identifying anomalous vertices. The package supports various graph packages (NetworkX, SGraph, iGraph, and GraphTools) and Machine Learning packages (SciKit and GraphLab). This project is under development and has a many planned improvements. More details on the project can be find in the our paper titled "Unsupervised Anomalous Vertices Detection Utilizing Link Prediction Algorithms" and in the NetSciX 2017 presentation.

Installation

git clone git://github.com/Kagandi/anomalous-vertices-detection.git
pip install -r requirements.txt
For Windows I suggest using conda and running:  run: conda env create -f environment.yml
The default installation only installs networkx and scikit-learn.
It is also possible to use the package with igraph, SGraph, GraphLab and GraphTools.
Note: Currently the package works best with networkx + scikit-learn or networkx + GraphLab.

Usage

Init:

from anomalous_vertices_detection.configs.graph_config import GraphConfig
from anomalous_vertices_detection.graph_learning_controller import *
from anomalous_vertices_detection.graphs.graph_factory import GraphFactory
from anomalous_vertices_detection.learners.gllearner import GlLearner

labels = {"neg": "Real", "pos": "Fake"}
dataset_config = GraphConfig("my_dataset", my_dataset_path, is_directed=True)
gl = GraphLearningController(GlLearner(labels=labels), dataset_config)
my_graph = GraphFactory().make_graph_with_fake_profiles(dataset_config.data_path,
                                            is_directed=dataset_config.is_directed,
                                            pos_label=labels["pos"], neg_label=labels["neg"])