Repository for Semantic-guided Machine Learning Algorithms

Knowledge Graph Constraints for Multi-Label Graph Classification

This is an implementation of the constraint-based subgraph pattern mining algorithm.

For details, check our DamNet'16 paper

The gspan module contains a modified version of the original gspan algorithm.

The following feature selection metrics are supported:

• Information Gain
• Top-k frequent
• GMLC (Kong et al. 2012)

You can add your own Must-Link and Cannot-Link constraints implementation in constraints.py

For multi-process evaluation, check multi_process_eval.py

Semantic Graph Kernel Lasso (GraKeLasso) (0.9)

This is a simple Python API for training and evaluating graph-regularized linear regression models

It was built to test the ideas of graph kernel regularization - see our ISWC'15 paper

To get started, have a look at testgrake.py

You need to provide data and the Laplacian matrix of the semantic graph as in data

Version 0.9

Most important features are:

• Loading data and regularization matrices
• Standard Lasso Coordinate-descent implementation
• Modified Coordinate-descent for graph-regularization implementation
• n-fold cross-validation

Requirements

• Python (>= 2.7)
• NumPy (>= 1.9)
• SciKit-Learn (>=0.15.2)
• Pandas (>=0.15.1)