💾 Source code and 📓 TeX for AL Hayes, M Das, P Odom, and S Natarajan, "User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams." Knowledge Capture Conference 2017
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
TeX_src
datasets
diagrams
media
.gitignore
LICENSE
README.md
walker.py
walker_demo.py

README.md



Source code and TeX for "User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams." K-CAP 2017

Citation

If you build on this code or the ideas of the paper, please use the following citation.

@inproceedings{kcap2017ermodeconstruction,
  author = {Alexander Hayes and Mayukh Das and Phillip Odom and Sriraam Natarajan},
  title  = {User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams},
  booktitle = {KCAP},
  year   = {2017}
}

Getting Started

Modes are used to restrict/guide the search space and are a powerful tool in getting relational algorithms such as BoostSRL to work. If your algorithm does not learn anything useful, then the first debug point would be the modes (in the background.txt file).

Walk-ER is a system for defining background knowledge for use in relational learning algorithms by exploring entity/attribute/relationships in Entity-Relational Diagrams. Refer to the BoostSRL Basic Modes Guide for more information about modes.

Prerequisites

  • Java 1.8
  • Python (2.7, 3.5)

Installation

  • Download the latest version from the GitHub repository (including five datasets):

    $ git clone https://github.com/batflyer/Walk-ER.git

Basic Usage

WalkER can either be invoked from a terminal or imported as a Python package. Examples of both follow:

  1. Interactive version:

    • Options overview (output of python walker.py -h):
    usage: WalkER_rewrite.py [-h] [-v] [--number NUMBER] [-w | -s | -e | -r | -rw] diagram_file
    
    positional arguments:
      diagram_file
    
    optional arguments:
      -h, --help         show this help message and exit
      -v, --verbose      Increase verbosity to help with debugging.
      --number NUMBER    Select number of features to walk to (assumes that
                         Important features are ordered from most important to
                         least important). Defaults to number_attributes +
                         number_relations if chosen number is greater than both.
      -w, --walk         [Default] Walk graph from target to features.
      -s, --shortest     Walk the graph from target to features. If there are
                         multiple paths, take the shortest. If the shortest are
                         equal lengths, walk both.
      -e, --exhaustive   Walk graph from every feature to every feature.
      -r, --random       Ignore features the user selected and walk (-w) from the
                         target to random features.
      -rw, --randomwalk  Walk a random path from the target until reaching a depth
                         limit (specified with --number).
    
    • Examples:

      • $ python walker.py -w diagrams/imdb.mayukh

      • $ python walker.py -rw --number 10 diagrams/imdb.mayukh

  2. As an imported package:

    import walker
    from boostsrl import boostsrl
    
    """Read a diagram file as a string."""
    with open('diagrams/imdb.mayukh') as f:
        diagram = f.read()
    
    """walk method accepts the same algorithm names as the interactive version."""
    bk = walker.walk(diagram, algo='w', n=3)
    target = [bk.target]
    
    """Use the boostsrl package to construct modes."""
    background = boostsrl.modes(bk, target, useStdLogicVariables=True, maxTreeDepth=4, nodeSize=3)
    
    ...
    

Acknowledgements

  • Mayukh Das and Sriraam Natarajan gratefully acknowledge the support of the CwC Program Contract W911NF-15-1-0461 with the US Defense Advanced Research Projects Agency (DARPA) and the Army Research Office (ARO).
  • Phillip Odom and Sriraam Natarajan acknowledge the support of the Army Research Office (ARO) grant number W911NF-13-1-0432 under the Young Investigator Program.
  • Icon in the logo is "Trail" by Martina Krasnayová from the Noun Project, used under a Creative Commons (CC) Attribution 3.0 United States License.