Scalable tensor factorization: Ext-RESCAL 0.7 is out

I've published Ext-RESCAL 0.7 on GitHub. The version has critical fixes in using proper sparse matrix types (SciPy module) for representing tensor slices. In my current experiments, I could handle a tensor (derived from DBpedia 3.9) of ~ 6 million x 6 million x 1000 dimensions, including over 180 million non-zero values, on ~ 70GB RAM. To the best of my knowledge, this is the largest dataset, for which RESCAL has been used so far.

RESCAL factorization (taken from [Nickel, 2013])

If you are interested in details of the problem, this software aims to solve, see the previous post on Machine Learning with Knowledge Graphs.

Fond of machine learning and remote control helicopters and RC quadcopters? Buy one on AliExpress!