zekarias-tilahun / mineral-python

Implementation of the Mineral algorithm as described in the paper, Mineral: Multi-modal Network Representation Learning.

Mineral (0.0.2)

Implementation of the Mineral algorithm as described in the paper, Mineral: Multi-modal Network Representation Learning. This version includes a minor change for cascade sampling, if observed cascades are provided. To enable this functionality just activate the --sample flag. Without this flag, it is exactly the first version Mineral 0.0.1.

Requirements!

• gensim 3.14
• Numpy 1.14+
• networkx 2.0+

Usage

Example usage

$ cd src
$ python mineral.py

Input Files

Graph inputs

Inputs related to the graph can be provided using --net-file and --att-file

Four kinds of formats are supported for both files, which are adjlist, edgelist, mattxt, and matnpy

adjlist

Format
node (list of neighbors | list of attributes)

Example
0 1 2 3
1 0 3
2 0
3 0 1

edgelist

Format
node (neighbor | attribute)

Example
0 1
0 2
0 3
1 0
1 3
2 0
3 0
3 1

mattxt or matnpy

Format
A matrix W of n x m dimensions

n is the number of nodes 
m is equal to n if W is an adjacency matrix of the graph
m is the number of attributes if W is an attribute matrix.
W can be a *.txt text or *.npy numpy binary file.

Cascade inputs

The final optional inputs are cascades, provided using --cas-file

Format:
(list of nodes)

Example:
2 0 1
1 0
0 2 1

Available command line options

--net-file: Path to a network file. Default is ../data/network.txt

--net-format: Network file format. Possible values are edgelist, adjlist, mattxt, and matnpy . Default is edgelist

--att-file A path to nodes attribute file. Default is ../data/attributes.txt

--att-format: Attribute file format. Possible values are edgelist, adjlist, mattxt, and matnpy. Default is mattxt

--cas-file: A path to cascades file. Default is ../data/cascades.txt

--sim-file: A path to save the simulated cascades. Default is '../data/simulated_cascades.txt'

--emb-file: A path to save the embedding files. Default is ../data/network.emb

--directed: A flag to indicate the input graph is directed. Default is False

--undirected: A flag to indicate the input graph is undirected. Default is True

--weighted: A flag to indicate the input graph is weighted. Default is False

--unweighted: A flag to indicate the input graph is unweighted. Default is True

--min-threshold: A threshold for the min length of observed cascades

--max-threshold: A threshold for the max length of observed cascades

--dim: An embedding dimension. Default is 128.

--window: A window size for the SkipGram model. Default is 10

--iter: The number of epochs. Default is 20

--r: The number of cascades to simulate from each node. Default is 10

--h: The maximum number of nodes to infect in a single simulation (length of each simulated cascade). Default is 60.

--workers: The number of parallel jobs. Default is 8

Citing

If you find Mineral relevant to your research, we kindly ask you to cite the paper:

@inproceedings{MOD17,
    booktitle="Proc. of the 3rd International Conference on Machine Learning, Optimization and Big Data",
    series = "MOD'17",
    author = {Kefato, Zekarias T. and Sheikh, Nasrullah and Montresor, Alberto},
    title = "Mineral: Multi-modal Network Representation Learning",
    month = sep,
    year = 2017,
    type = {CONFERENCE},	
    publisher = {Springer},
}