This project implements P. Talukdar's Modified Adsorption (MAD) label propagation algorithm. MAD is a graph semi-supervised learning model. As such it uses a small number of seeds to determine the class. It is transductive, it assigns classes to the already existent dataset. It does not assign classes to new, unseen instances.
The module includes a simple class implementing the MAD algorithm. A graph file is needed, graph_file, and a seed_file, containing the minimum set of initial labeled items. They must be in a tab separated format: NodeA, NodeB, LinkWeight. For example, the graph_file should be like this:
N1 N2 0.18
For the seed_file, a Node, a Label and a Weight (indicating the strength of the label class):
N1 L1 1.0
N4 L2 1.0
Then, it suffices to call the MAD constructor:
mad = ModifiedAdsorption(graph_file, seed_file)
Calculate the modified adsorption:
mad.calculate_mad()
And finally, get the results:
mad.results()
The code is not thoroughly tested, the matrices are not checked for special considerations neither at the beginning of the process, neither at the iterative steps.
I am sure it could be made faster and more memory efficient. I have tested it with around 10k nodes and it took around 1hr to converge. I need to perform more experiments, specially take time to run these datasets: http://www.talukdar.net/datasets/class_inst/
Partha Pratim Talukdar and Koby Crammer. 2009. New Regularized Algorithms for Transductive Learning. In Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases: Part II (ECML PKDD '09), Wray Buntine, Marko Grobelnik, Dunja Mladenić, and John Shawe-Taylor (Eds.). Springer-Verlag, Berlin, Heidelberg, 442-457. DOI=10.1007/978-3-642-04174-7_29 http://dx.doi.org/10.1007/978-3-642-04174-7\_29
Apache v2