Thebeast System

From GM-RKB
Jump to navigation Jump to search

thebeast System is a Markov Logic Software System.



References

  • http://code.google.com/p/thebeast/
    • markov thebeast is a Statistical Relational Learning software based on Markov Logic.
    • It allows you to implement relational learning and structured predictions problems such as
      • Entity Resolution
      • Link Prediction
      • Dependency Parsing
      • Semantic Tagging
      • Sentence Compression
      • Chunking
    • by simply defining a model and providing training data. Learning and inference are both handled by thebeast.
    • thebeast uses Markov Logic as language to describe complex Markov Networks. In Markov Logic first order predicates and formulas describe the nodes and connectivity of Markov Networks. Compared to alchemy, another Markov Logic Engine, thebeast uses a different MAP inference technique (Integer Linear Programming with Cutting Planes), and may be faster at times.
    • Features
      • Fast and exact MAP inference using Cutting Planes combined with Integer Linear Programming (or Max-Walk-Sat), see the corresponding paper on Cutting Plane Inference for Markov Logic
      • Online Discriminative Training using MIRA, Perceptron and Passive Aggressive Learning
      • Parametrized weights: use one formula to describe millions of features with individual weights
      • Shell: Use a interpreter to edit and inspect your model. Allows you to analyze errors, inspect feature weights and scores
      • Cardinality constraints: make statements about how often certain conditions should hold in a solution.