Description

This web page presents the RelSets algorithm for finding closed sets for labeled data. Closed sets for labeled data characterize the space of relevant combinations of features for discriminating the target class. In practice, the result of RelSets is a set of non-redundant rules for describing the target class.

The web service for closed sets for labeled data (RelSets for short) is available at http://zulu.ijs.si:8085/RS_services?wsdl. RelSets mining can also be run as a web application from the link below.

Web application

The web application for closed sets for labeled data (RelSets for short) is available at http://zulu.ijs.si/WEB/RelSets/.

Usage

A sample code for running the service in the Python programming language is here.
the Python script: testRelSets.py
the data: lenses.tab

The testRelSets.py script is shown here:

# generate proxy
from suds.client import Client
client = Client('http://zulu.ijs.si:8085/RS_services?wsdl')

# load the data
data = open('lenses.tab').read()

# call service
response = client.service.relSets(inputTable=data, minTPrate=0.8)

# print results
print response.rulesAsString     # return rules as human-readable text
print response.rulesAsPMML       # return rules as PMML text

	  

Data

The web service takes as the input the data in the Orange data format. The target variable needs to be nominal (in Orange terminology: discrete). Look at the Orange web page to find more details and sample datasets. The Orange native tab format is recommended.

Sample output

The RelSets web service provides two types of output: one is in the PMML format and one in a human-friendly text format. Both are illustrated below.

PMML output: sample_output.pmml

Human-friendly output:

quality= 1.60 complexity= 1 covered=12 numTP= 8 numFP= 4  conf= 0| astigmatic = yes  -> lenses = none
quality= 1.60 complexity= 1 covered=12 numTP= 8 numFP= 4  conf= 0| prescription = hypermetrope  -> lenses = none
quality= 12.00 complexity= 1 covered=12 numTP=12 numFP= 0  conf= 1| tear_rate = reduced  -> lenses = none
quality= 1.50 complexity= 0 covered=24 numTP=15 numFP= 9  conf= 0|   -> lenses = none
quality= 2.00 complexity= 3 covered= 2 numTP= 2 numFP= 0  conf= 1| tear_rate = normal   age = young   astigmatic = yes  -> lenses = hard
quality= 3.00 complexity= 3 covered= 3 numTP= 3 numFP= 0  conf= 1| tear_rate = normal   prescription = myope   astigmatic = yes  -> lenses = hard
quality= 1.33 complexity= 2 covered= 6 numTP= 4 numFP= 2  conf= 0| tear_rate = normal   astigmatic = yes  -> lenses = hard
quality= 3.00 complexity= 3 covered= 3 numTP= 3 numFP= 0  conf= 1| tear_rate = normal   prescription = hypermetrope   astigmatic = no  -> lenses = soft
quality= 2.50 complexity= 2 covered= 6 numTP= 5 numFP= 1  conf= 0| tear_rate = normal   astigmatic = no  -> lenses = soft

Publications

Gemma C. Garriga, Petra Kralj, Nada Lavrač
Closed Sets for Labeled Data
Journal of Machine Learning Research, 9(Apr):559-580, 2008.
[ps] [pdf] [bib]

Gemma C. Garriga, Petra Kralj, Nada Lavrač
Closed Sets for Labeled Data
In Proceedings of the 10th European Conference on Principles and Practice of Knowledge Discovery (PKDD 2006), Springer, ISBN 978-3-540-45374-1, year 2006, pages 163-174.
[ps] [pdf] [bib]

Petra Kralj, Ana Rotter, Nataša Toplak, Kristina Gruden, Nada Lavrač, Gemma C. Garriga
Application of closed itemset mining for class labeled data in functional genomics
Informatica Medica Slovenica, ISSN =1318-2129, year 2006, number 1, pages 40-45.
[ps] [pdf] [bib]