Ingo Thon's Website

Datasets used in my papers

Stochastic relational processes: Efficient inference and applications

If you use one of the following models or datasets please cite:
	  @article{thon10srp,
	  	  author  = { Ingo Thon and Niels Landwehr and Luc {De Raedt}},
	  	  title   = { Stochastic relational processes:
	  	              Efficient inference and applications           },
	  	  year    = { 2010},
	  	  journal = { Special issue of Machine Learning Journal
	  	              on Mining and Learning with Graphs              },
	  	  editors = { {S V N} Vishwanathan and Samuel Kaski and
	  	  	      Jennifer Neville and Stefan Wrobel              }
	  }
	

Stochastic Blocks World Domain

This domain is a stochastic version of the well-known artificial blocks world domain, representing an agent that is moving blocks which are stacked on a table. We use this artificial domain to perform controlled experiments, testing the scaling and convergence behavior of inference and learning algorithms.

Model (Comming soon)
No Dataset, sampled on demand.
Description

Chat Room Domain

This domain is concerned with the analysis of user interaction in chat rooms. We have monitored a number of IRC chat rooms in real time, and recorded who was sending messages to whom using the PieSpy utility. This results in dynamically changing graphs of user interaction, representing the social network structure among chat room participants. We learn these dynamics using separate models for different chat rooms. The resulting set of models can be used to visualize commonalities and differences in the behavior displayed in different chat rooms, thereby characterizing the underlying user communities.

Model
Dataset 2009/06/09 (additional days will be added later)
Description

Massively Multiplayer Online Game Domain

As a final evaluation domain, we consider the large-scale massively multiplayer online strategy game Travian. Game worlds feature thousands of players, game artifacts such as cities, armies, and resources, and social player interaction in alliances. Game states in Travian are complex and richly structured, and transitions between game states highly stochastic as they are determined by player actions. We have logged the state of a ``live'' game server over several months, recording high-level game states. We address different learning tasks in the Travian domain, such as predicting player actions (prediction setting) and identifying groups of cooperating alliances (classification setting).

Model
Dataset
Description
For this dataset please also consider to cite the following paper
Boosting Relational Sequence Alignments (Andreas Karwath, Kristian Kersting, and Niels Landwehr)

Don't Fear Optimality: Sampling for Probabilistic-Logic Sequence Models

Comming soon...