leoribeiro / struc2vec Goto Github PK

View Code? Open in Web Editor NEW

379.0 13.0 115.0 19.31 MB

This repository provides a reference implementation of struc2vec.

License: MIT License

Python 100.00%

graph algorithm

struc2vec's Introduction

struc2vec

This repository provides a reference implementation of struc2vec as described in the paper:

struc2vec: Learning Node Representations from Structural Identity.
Leonardo F. R. Ribeiro, Pedro H. P. Saverese, Daniel R. Figueiredo.
Knowledge Discovery and Data Mining, SigKDD, 2017.

The struc2vec algorithm learns continuous representations for nodes in any graph. struc2vec captures structural equivalence between nodes.

Before to execute struc2vec, it is necessary to install the following packages:
pip install futures
pip install fastdtw
pip install gensim

Update

Python 3 version: https://github.com/sebkaz/struc2vec/tree/master

Basic Usage

Example

To run struc2vec on Mirrored Zachary's karate club network, execute the following command from the project home directory:
python src/main.py --input graph/karate-mirrored.edgelist --output emb/karate-mirrored.emb

Options

To activate optimization 1, use the following option: --OPT1 true
To activate optimization 2: --OPT2 true
To activate optimization 3: --OPT3 true

To run struc2vec on Barbell network, using all optimizations, execute the following command from the project home directory:
python src/main.py --input graph/barbell.edgelist --output emb/barbell.emb --num-walks 20 --walk-length 80 --window-size 5 --dimensions 2 --OPT1 True --OPT2 True --OPT3 True --until-layer 6

You can check out the other options available to use with struc2vec using:
python src/main.py --help

Input

The supported input format is an edgelist:

node1_id_int node2_id_int

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 struc2vec.

Miscellaneous

Please send any questions you might have about the code and/or the algorithm to [email protected].

Note: This is only a reference implementation of the framework struc2vec.

struc2vec's People

Contributors

Stargazers

Watchers

Forkers

cztgit benjamesbabala zwytop redsuncmx jiajiadf anryyang tandychao lookfwd bill-codes chaitanyacixlive meccy shyant zhuiyuan616124 embedxj swalter2 liwzhi aabbcc0812206523 uotter daniel-tan youngleec ntuaha songfgh ai3dvision syd951186545 jalamao steenfatt afcarl xuequeen zhouchunpong torix yuehanlyu zbn123 joisino penngao xdlequ dreampppaper mafm08 mazelinx heyuanhao fcytrina caiocrr xuyou314 tianjiansmile jacopoorlandini 30lm32 rcastroab yannan1212 cgoliver ymitiku summerrain1994 guyulongcs hins naihemeng miliana hoangdzung jjh3024 saberse nwilliam868 toatmosto tsingh-googl vivian457 shaydeu cslele coolsunxu catezi dsysoev rain0831 treenewbee0 janeyzy abel0828 synergetic-ai zjzijielu elvissmog liun-online junwei-h mzy2240 zhenglinyi maxmit233 yang0110 moguizhizi hitsongxiao jumxglhf world4jason baobunuo fanyang1997 blakedallen patagona-technologies yiran-hao dmitrijsk 748978460 ypman tiger-tiger tuantx7110 lengfeng343 hxynl w55699 bzp92 saisaitian wheat-ops libertaspatronus

struc2vec's Issues

a problem about Barbell graph

Thank you for your code!
i use the file struc2vec/emb/barbell.emb provided by your code to draw the latent representations,but i can't get the result in your paper figure2(e)
Look forward to your answer and thanks again!

Python3

The current solution is with Python2. Is there any python3 version of codebase available?

RuntimeWarning: invalid value encountered in double_scalars

When i run the code with Cora dataset ,the following occurs:
/home/songzenghui/zee/struc2vec/src/algorithms_distances.py:541: RuntimeWarning: invalid value encountered in double_scalars e_list = [x / sum_w for x in e_list]
The dataset has 2708 nodes,5429 edges,and i ran python src/main.py --input graph/cora.edgelist --output emb/cora.embeddings --OPT1 True --OPT2 Ture --OPT3 True
Is the sum_w too small so that the division is invalid?

A question about labels-europe-airports.txt

What is the meaning of this label?
Thanks in advance!

TypeError: 'dict_keys' object is not subscriptable

Has anybody come across this error:

I've tried converting the package to be compatible with Python 3, but still left with this error - not sure where to start to fix it.

Thanks

struc2vec on weighted graph

The arguments args.weighted and args.unweighted are never passed in any module, and the default argument is unweighted = True. Do the weighted/unweighted arguments have any effect?

A license

What license does this reference implemention fall under? MIT?

issue attribute

Traceback (most recent call last):
  File "src/main.py", line 128, in <module>
    main(args)
  File "src/main.py", line 121, in main
    G = exec_struc2vec(args)
  File "src/main.py", line 97, in exec_struc2vec
    G = struc2vec.Graph(G, args.directed, args.workers, untilLayer = until_layer)
  File "/Users/anton/Downloads/struc2vec-master/src/struc2vec.py", line 20, in __init__
    self.G = g.gToDict()
  File "/Users/anton/Downloads/struc2vec-master/src/graph.py", line 127, in gToDict
    for k,v in self.iteritems():
AttributeError: 'Graph' object has no attribute 'iteritems'

Problem with disconnected nodes

I work with some graphs that have some disconnected nodes. I believe that when I create the edges list, which contains only the connected part of the network, the algorithm outputs network representations for only the connected part of the network. Is there any way to include the disconnected nodes, or I should remove them from the network, completely?

ask fo help

Karate-mirrored.edgelist : what are two columns represented? Hope to get your answer，thx

filr

Hello,I can not find emb.karate.emb data,whether you forget put it

Node attributes/labels

If struc2vec able to handle cases with node attributes/labels?
Or is it reasonable to incorporate node attributes/labels with struc2vec?

The problem about the scalability

I tried to use the struc2vec to train on 1 million nodes with 24 threads,but no embeddings was generated after three days.Is there a solution?

plot the embedded nodes

Hi, thanks for your open-source code firstly. But, I have a question: I use t-sne to plot the embedded 128 dimensions vectors in 2D space, It's distribution seem not be similar as the paper with karate dataset figure, so I want to ask if you are using t-sne to visulization? thank you for your kindness!