Skip to content

README.md

Latest commit

 

History

History
65 lines (52 loc) · 2.47 KB

README.md

File metadata and controls

65 lines (52 loc) · 2.47 KB
Raw

Unsupervised baselines (12 files)

This folder contains the unsupervised baselines that the supervised method is compared against, Common Neighbor (CN) and Node-Link Clustering (NLC) ranked top-L missing-link prediction.

Run

From the repository root:

addpath(genpath('Unsupervised'));
addpath('data');
cd Unsupervised

Unsupervised_CN_V1     % CN baseline
% or
Unsupervised_NLC_V0    % NLC baseline (Wu et al. 2016)

Both scripts:

  1. Build the adjacency matrix for the selected dataset.
  2. Randomly hide 10 % of the existing edges as test edges.
  3. Score every non-existing edge in the training graph using the chosen similarity index.
  4. Take the top-L scored pairs as predicted links.
  5. Report Accuracy, Sensitivity, Specificity, Precision, Recall, F-measure, G-mean and AUC (averaged over 10 independent runs).

The top-L cut-off and the dataset are configured at the top of each script:

L_top      = 100;
num_of_run = 10;
dataset    = 'Jazz';

File map

File Role
Unsupervised_CN_V1.m CN-based top-L link prediction
Unsupervised_NLC_V0.m NLC-based top-L link prediction (Wu et al. 2016)
adj_gen.m dataset loader (looks in ../data/)
CN.m Common Neighbor scoring
NLC.m Node-Link Clustering scoring (Wu et al. 2016)
ACO_Feature_Extract.m also present here so the folder can run a quick PH-based unsupervised score as an extra baseline; differs slightly from the Supervised/ version
main_feature_extraction.m older, exploratory single-run driver
Conv2EdgeList.m converts an N×N feature matrix to a per-edge list
Evaluate.m confusion-matrix metrics (Accuracy / Sensitivity / Specificity / Precision / Recall / F-measure / G-mean)
roc_curve.m ROC curve and AUROC
sort_column1.m descending sort by column 1 (NB: in Supervised/ the same-named file sorts ascending — do not mix folders on the MATLAB path)

Why are several filenames shared with Supervised/?

adj_gen.m, CN.m, NLC.m, Conv2EdgeList.m, Evaluate.m, roc_curve.m, sort_column1.m, ACO_Feature_Extract.m, and main_feature_extraction.m all appear in both folders. Most are identical, but a few differ in non-trivial ways, most importantly sort_column1.m, whose sort direction is flipped between the two folders to suit each pipeline. Keep the two folders separate on the MATLAB path; do not addpath both at the same time unless you know which version will be shadowed.