Dazzle

Download

Download the latest Windows version: DazzleSetup.exe (installer), dazzle.zip (zip containing only the executable).

News

October 19, 2005: The XTC library (eXtended & Typed Controls for wxHaskell) is available for download: XTC.hs, documentation.

July 19, 2005: The paper "Haskell Ready to Dazzle the Real World" (pdf, bib) has been accepted to appear on the Haskell Workshop.

May 18, 2005: new version with

  • Compute sensitivity and specificity in confusion matrix
  • Change the order of values
  • New file format (XML)
  • Bugfix: having minimized windows at application close no longer moves these windows out of screen at the next startup
  • Sensitivity analysis (not finished yet)

April 12, 2005: new version with:

  • Substitute 0 by some user-specific number in the probability tables of a learned classifier
  • Confusion matrix also shows accuracy
  • Ten-fold cross validation: generates 10 confusion matrices with accuracies and computes average accuracy
  • Window positions and sizes are remembered between sessions
  • Windows can be made smaller than before

February 28, 2005: new version with:

  • Loopy propagation
  • Updated user manual
  • Test selection experiments
  • More intuitive user interface
  • Case browser and data analyser have been merged
  • Fix for saving a file to a full disk
  • Compute relation in data browser: create a probability table based on a data set
  • Select nodes for viewing inside the data browser
  • Nodes that are shown in posteriors pane are saved to disk
  • Floating-point numbers are rounded to 6 decimals
  • Numbers in posterior charts are rounded to 1 decimal
  • Several bug fixes

January 20, 2005: new version with:

  • numbers in the posteriors charts
  • confusion matrix
  • learning classifiers
  • lots of small improvements

December 7, 2004: new version with unlimited undo/redo in network and case browser; export probability tables

About

Dazzle is a tool for editing and analysing Bayesian networks and is being developed at the decision support group of the institute of information and computing sciences of Utrecht University.

Implementation

Dazzle is implemented in the Haskell programming language. Visit the Haskell website for more information on this powerful language.

For probabilistic inference we use the efficient SMILE framework of the Decision Systems Laboratory (University Of Pittsburgh). From the website

SMILE [Structural Modeling, Inference, and Learning Engine] is a fully platform independent library of C++ classes implementing graphical probabilistic and decision-theoretic models, such as Bayesian networks, influence diagrams, and structural equation models.

Visit the SMILE website website library for more information.

Features

Screenshot on Windows XP

  • Build networks with few clicks
  • Import and export networks in Hugin (.net) and Genie (.dsl) format
  • Allow cycles while designing the network
  • Specify dimensions of network in centimeters (useful for including networks in a paper)
  • Unlimited undo and redo in both the network editor and the case browser
  • Names with funny characters are supported in node names and value/state names, e.g. "5<x<10" as a value name
  • State-of-the-art test selection (research only at the moment)
  • Labels can appear above and below nodes
  • Case browser (see screenshot)
    • Look at separate cases showing input (evidence) and output (posteriors)
    • Vertical bar charts for posteriors you are interested in
    • Probabilities as numbers with three significant digits above the bars
    • Add and delete cases
    • Open and save case files
    • Compute confusion matrix
  • Probability table window
    • View and edit probability tables without opening dialogs
    • Probabilities that have not been entered yet are indicated by a minus as opposed to inventing some distribution as other tools do
    • Save networks with incomplete tables
    • Save probability table(s) to disk
  • Data browser
    • Learn classifiers from data
    • Naive, TAN and k-DB classifiers can be learned
    • Feature selection: backward elimination, forward selection or a full classifier
    • Compute the relation between a node and some (virtual) parents w.r.t. the data set
    • Compute confusion matrices
    • Compute the MDL score of a network
  • Logic sampling
  • Print networks

Contact

Arjan van IJzendoorn (afie@cs.uu.nl)
Martijn Schrage (martijn@cs.uu.nl)