RockIt: A query engine for Markov logic RockIt answers maximum a-posteriori (MAP) queries (also called MPE queries), marginal queries (also called probability queries), and learns the weights of Markov logic networks. Markov logic is a combination of Markov networks and first-order logic. RockIt is currently the fastest MAP query engine for Markov logic networks.

Maximum A-Posteriori (MAP) Inference

RockIt compiles MAP queries to integer linear programs. During this translation we apply the meta algorithms cutting plane inference (CPI) and cutting plane aggregation (CPA). In CPI we only add the constraints which are violated and solve several smaller ILPs until no new violated constraints are found. This usually leads to much smaller ILPs and, thus, faster run times. In CPA we leverage symmetries in the Markov logic network by aggregating ground clauses. This reduces the required number of constraints and variables in the ILP translation and makes symmetries more explicit to state-of-the-art ILP solvers. Internally, RockIt utilizes in-memory databases from MySQL and the integer linear solver Gurobi (a free academic license is available).

Information

• Google code: https://code.google.com/p/rockit/
• Web service: http://executor.informatik.uni-mannheim.de/systems/rockit/