The mirand algorithm learns continuous representations for nodes in any (un)directed, (un)weighted graph. Paper accepted at ECAI-2020 (http://ecai2020.eu/papers/1648_paper.pdf)

Environment Set-up

Use python version 2.7

• Clone the repository.
• Navigate to the base directory of mirand (the download location)
• Create a virtual environment using the following command:
  virtualenv venv
  (If virtualenv package is not installed, please install using pip)
• Activate the environment using:
  source venv/bin/activate
• Install required python modules to run the code.
  pip install -r requirements.txt

Congratulations!! You are now setup to run the code.

Basic Usage

Input

• Look at the sample dataset cora (residing inside data directory). If you want to experiment on different datasets, create a folder with name of your dataset.
• Two files are required to run and generate the embedding - edgelist file for structure graph and edgelist file for content graph
• Naming convention for link structure layer: <dataset_name>_struc.edgelist
• Naming convention for content/attribute layer: <dataset_name>_attr.edgelist

Example

To run mirand on cora network, execute the following command from src directory inside the project home path:
python main.py --input-struc ../data/cora/cora_struc.edgelist --input-attr ../data/cora/cora_attr.edgelist --output ../data/cora/cora.embed --dataset=cora --dimensions=128

Options

You can check out the other options available to use with mirand using:
python src/main.py --help

Input

The supported input format is an edgelist:

node1_id_int node2_id_int <weight_float, optional>

The graph is assumed to be undirected and unweighted by default. These options can be changed by setting the appropriate flags.

Output

The output file has n+1 lines for a graph with n vertices. The first line has the following format:

num_of_nodes dim_of_representation

The next n lines are as follows:

node_id dim1 dim2 ... dimd

where dim1, ... , dimd is the d-dimensional representation learned by mirand.