The WinMine Toolkit

1
The WinMine Toolkit

• Max Chickering

2
Build Statistical Models From Data

• Dependency Networks
• Bayesian Networks
• Local Distributions
• Trees
• Multinomial / Binary Multinomial
• Gaussian / Binary Gaussian
• Log Gaussian / Binary Log Gaussian
• Complete Tables

3
Data Processing Tools

• DataConverter.exe (Interactive)
• Convert raw text or SQL data into XML format
• DataCheck.exe (Command-line)
• Extract basic statistics from data
• DataJoin.exe (Command-line)
• Perform a join between two datasets
• DataSplit.exe (Command-line)
• Split data into train/test

4
Modeling Tools

• PlanEditor.exe (Interactive)
• Specify roles (e.g. input vs output) and
• distributions for variables
• Dnet.exe (Command line)
• Build a dependency network or Bayesian
• network from data
• DnetBrowser.exe (Interactive)
• Interactively browse dependency network or
• Bayesian network
• DnetLogscore.exe (Command Line)
• Evaluate Prediction accuracy of models

5
Built-In Help -help Option
c\WinMine Toolkit\Bingtdatacheck -help This
executable parses a data file and prints
out summary statistics. If a marginal statistics
file is provided with the '-marg' flag, the
executable collects marginal counts for each
variable and prints them to that file.
6
Built-In Help No Arguments
c\WinMine Toolkit\Bingtdatacheck Error in command
line required argument '-data' not given syntax
for datacheck Flag Type Description
Optional? Default ----------------------
--------------------------------------- -data
string Data file no -report
string Report file yes -marg
string Marginal counts file yes -silent
bool Suppress progress output yes
false -help bool Display help
yes false c\WinMine Toolkit\Bingt
7
Interactive Mode
c\WinMine Toolkit\Bingtdatacheck -gui
8
WinMine Home Pagehttp//www.research.microsoft.c
om/dmax/WinMine/Tooldoc.htm

• Download/Update tools
• No registry changes simply copies executables
• Online Tutorial
• Steps through using all of the tools with a
  simple example
• Discussion Group