Markov Logic Network

Tutorials | Documentation | Release Notes | Roadmap

MLN is a distributed Markov Logic Network in Julia, inspired by the works of Pedro Domingos, Stanford's Tuffy and Felix, Hélène Papadopoulos, and a few others.

MLN is a unified learner, and can learn broadly or narrowly from any application.

Key points will be listed after some code is done.

Installation

Hello World

Inputs, in three parts

1. MLN program

   • Predicate definitions (constants and variables) (user beliefs) (optional)

   • e.g. Zika reported and unreported cases increased after the 2016 Olympics

   • Rule definitions (constants and variables with weights, 0.0-1.0) (optional)

   • e.g. Belief is 0.9 that it did

2. Evidence (observed constants and variables) (optional)

   • e.g. US, Hawaii, Canada, Cuba, Puerto Rico, Haiti, Brazil, Poland, Singapore cases increased

3. MAP Query (user questions about a closed or open world)

   • e.g. Did Zika cases, reported and not, increase for all countries?

   Marginal Query (user questions about an element within a world)

   • e.g. Did Zika unreported cases in Greenland increase?

Outputs, of two types

1. MAP inference (broad learning)

   • e.g. Reported and unreported Zika cases increased for countries x, y, z

2. Marginal inference (narrow learning)

   • e.g. Unreported Zika cases in Greenland increased by a likelihood of 64%