henhoe2vec 1.0.4

Implementation of the HeNHoE-2vec algorithm by Valentini et al. (2021).

HeNHoE-2vec

A Python implementation of the HeNHoE-2vec algorithm by Valentini et al. for the embedding of networks with heterogeneous nodes and homogeneous edges (HeNHoE).

Note: HeNHoE networks are analogous to multilayer networks: in HeNHoE networks, each node has a distinct node type, and in multilayer networks, each node belongs to a distinct layer. The terms type and layer may therefore be regarded synonymous. Throughout the code and for the remainder of this documentation, we will use the terms multilayer network and layer as opposed to HeNHoE network and type.

Installation

Install the package from PyPI by running the following command:

$ pip install henhoe2vec

Alternatively, clone this repository by running

$ git clone [e-mail]:Bertr0/HeNHoE-2vec.git

and then install the package by running pip install . from the root of the repository.

Usage

This package may be used as a Python script or as a package, allowing its modules to be imported by other Python projects. Both forms of use make it easy to run HeNHoE-2vec on multilayer networks.

As a Package

After installing the package using pip, its modules may be imported using

import henhoe2vec

The many individual steps of HeNHoE-2vec are accumulated in a single run() method in the henhoe2vec.henhoe2vec module. HeNHoE-2vec can be run from start to finish as follows:

import henho2vec as hh2v

hh2v.henhoe2vec.run(input_csv, output_dir)

input_csv is the path to the multilayer edge list of the network to be embedded (csv file with no index). output_dir is the path to the output directory where the embedding files will be saved. The run() method takes a bunch of other optional parameters which can be used to configure HeNHoE-2vec. A comprehensive overview of parameters can be found in the code documentation.

As a Python Script

To run HeNHoE-2vec as a script, clone this repository using

$ git clone [e-mail]:Bertr0/HeNHoE-2vec.git

, install the requirements found in requirements.txt and run the following command from the root of the repository:

$ python3 -m src.henhoe2vec --input <input_path> --output_dir <output_dir_path>

This will generate node embeddings for the nodes of the network specified by the multilayer edge list saved at <input_path> and saves the embedding files in <output_dir>.

Run python3 -m src.henhoe2vec --help from the root of the repository to show an overview of all arguments taken by the script. The following table also shows an overview of all arguments:

Script Arguments

Argument	Type	Description	Default Value
--input	str	Path to the multilayer edge list of the network to be embedded (csv file with no index).	-
--sep	str	Delimiter of the input csv edge list.	"\t"
--header	store_true	Pass this argument if the input csv edge list has a header.	-
--output_name	str	Name of the output .csv file (without suffix).	"embeddings"
--is_directed	store_true	Pass this argument if the network is directed.	-
--edges_are_distances	store_true	Pass this argument if edge weights indicate distance between nodes (opposed to weight/similarity).	-
--output_dir	str	Path of the output directory where the embedding files will be saved.	-
--dimensions	int	The dimensionality of the embeddings.	128
--walk_length	int	Length of each random walk.	20
--num_walks	int	Number of random walks to simulate for each node.	10
--p	float	Return parameter p from the node2vec algorithm.	1.0
--q	float	In-out parameter q from the node2vec algorithm.	0.5
--s	float	Default switching parameter for layer pairs which are not specified in the --s-dict argument.	1.0
--s_dict	list	Switching parameters for specific layer pairs in a dict-like manner. Pass the names of layer pairs followed by their switching parameters, separated by white spaces. E.g., if the switching parameter from layer1 to layer2 is 0.5 and the switching parameter from layer2 to layer1 is 0.7, you would pass layer1 layer2 0.5 layer2 layer1 0.7. Note that layer pairs are directed. For all layer pairs which are not specified here, the default parameter --s is adopted.	empty list
--window_size	int	Context size for the word2vec optimization.	10
--epochs	int	Number of epochs in SGD.	1
--workers	int	Number of parallel workers (threads).	8