Alchemy-2 extends Alchemy (suite of inference/learning algorithms for Markov Logic) by implementing a number of lifted inference algorithms. Lifted inference algorithms differ from traditional propositional inference algorithms by performing inference at the first-order level as far as possible and propositionalizing only as needed. Lifted Inference algorithms therefore offer far greater scalability when compared to propositional algorithms.
Magician controls inference complexity and scales-up to large domains by combining lifted inference with approximations and advanced solution counters. The current software is a Beta-version and implements Gibbs sampling along with learning using Contrastive Divergence. Please consider citing the papers: Venugopal et al. AAAI 2015 and Sarkhel et al. AAAI 2016, if you use this software for your research.