beliefnetdata File Reference

---

Detailed Description

Creates a data set by sampling from a Bayesian Network.

>This program creates a synthetic data set by loading an existing belief network (in BIF format) and sampling from it, possibly introducing noise. This program will also create a .names file for the resulting data set. More specifically, belifnetdata reads the network, generates samples from it with BNGenerateSample, and adds noise to them (if requested) with ExampleAddNoise.

Multiple runs with the same seed parameter produce the same results. Also note that running this command with one level of -v will output some statistics about the belif net which you might find useful.

VFML comes with a collection of benchmark belief nets, and you may want more information on these.

Arguments

• -f 'stem name for output'
  • (default DF)
• -bnf 'file containing belief net'
  • (default DF.bif)
• -train 'size of training set'
  • (default 10000)
• -infinite
  • Generate an infinite stream of training examples, overrides -train flags, only makes sense with -stdout (default off)
• -stdout
  • Output the trainset to stdout (default to 'stem>'data)
• -seed 'random seed'
  • (default to random)
• -noise 'noise level'
  • 10 means change 10% of values, including classes (default to 0)
• -target 'dir'
  • Set the output directory (default '.')
• -v
  • Increase the message level
• -h
  • Run with this argument to get a list of arguments and their meanings.
    

Example

beliefnetdata -f train -bnf alarm.bif -train 1000 -seed 111 -noise 5

Creates 1000 samples from the alarm network, randomly corrupts 5% of their values, write the resulting samples to train.data (and create a file train.names) and reproduce the same data set everytime the same arguments are used (thanks to the seed parameter)

---

Generated for VFML by hosted by