EDGE‎ > ‎

Manual

The pdf version of the manual can be downloaded at the end of this page in the download section.


EDGE

EDGE is an algorithm for learning the parameters of ontologies. It applies EM and computes the expectations by traversing twice the BDDs that are built for inference.


The system can take as input files defined in various OWL formats, including RDFXML. It

can compute the probability of all the queries that are allowed by Pellet explain function.


EDGE takes as input one or two RDFXML files for the ontology, one file for negatvive examples and one file for positive examples. For the ontology in case of single file, it contains

the ontology and the annotations for the axioms too. In case of two files, the first file contains the

full ontology and the second one contains all the annotations. In this case, the annotations contained

in the first file are not used.

Arguments

EDGE's parameters can be setted in two ways, via file or via command line interface.

Following we show all the parameters that can be setted.

File

In file all parameters is specified by <name_param>:<value>. The character '#' is used for comment.

  • positive    positive examples
  • negative   negative examples
  • ontology   ontology file
  • probability probability file, if not specified EDGE will use the same file of the ontology
  • randomize
    • true -> randomly set the probability of axioms with probability annotation
    • false -> EDGE uses the probability contained in the file
  • all -> ramdomly set the probability of all axioms (also the ones without probability annotation)
  • random_seed seed for random function
  • diff_ll max difference between log likelihood
  • ratio_ll max ratio between log likelihood
  • max_iter max number of iteration for the EM algorithm
  • max_expl max number of explanations for the queries
  • timeout max time allowed for the inference, 0 for unlimited time.
    • Format: [Xh][Ym][Zs][Kms] (Default: 0)
  • show_all true -> show all the axiom in output
  • test true -> run test
  • output
    • filename-> name for output file
    • none -> no output
  • format specify the format for output file, it can be:
    • DEFAULT (same as OWLXML)
    • OWLXML
    • OWLFUNCTIONAL
    • RDFXML

command line interface

use

BDDEM [arguments] <file_pos> <file_neg> <file_onto> <file_prob>

  • <file_pos> fileURI URI of positive examples file
    • -nopositive run without positive examples
  • <file_neg> fileURI URI of negative examples file
    • -nonegative run without negative examples
  • <file_onto> fileURI URI of ontology file
    • <file_prob> fileURI URI of probability file
    • -noprob / -noprobability for specify that the probabilities are contained in ontology file
    [arguments]
  • -r to randomize the starting probabilities
  • -rs <seed> to set the seed of randomize
  • -ra to randomize all the axiom in the starting ontology
  • -dll <value> to set the stop difference between log-likelihood of two consecutive EM cycles
  • -rll <value> to set the stop ratio between log-likelihood of two consecutive EM cycles
  • -it <value> to set the maximum number of cycles
  • -m / -m_ex <value> to set the maximum number of explanations to find for each query
  • -t to set the test mode
  • -to / -time / -timeout <value> to set the max time allowed for the inference, 0 for unlimited time
    • <value> format:  [Xh][Ym][Zs][Kms] (Default: 0)
  • -s / -sa to force the visualization of all results
  • -out <file_URI> to specify the file containing the learned parameters
  • -outf <format> to specify the format of the output file
    • <format> can be:
      • DEFAULT
      • OWLXML
      • OWLFUNCTIONAL
      • RDFXML 


Ċ
Riccardo Zese,
Dec 16, 2013, 7:16 AM
Comments