Picturing Interactively Statistically Sensible
Association Rules Reporting Overviews
Pissarro is a project of Dept. of Computer Oriented Statisics and Data Analysis (COSADA) at the Augsburg University, Germany.
Landscape at Chatonoval (Pissarro 1880)
Pissarro is an interactive Data Mining Software to calculate association rules from binary datasets.
Pissarro supports different algorithms for rulecalculation. Complex structures in real data implicate different algorithms for reasons of performance. To control quality and quantity of desired rules, minimum support, minimum confidence and a maximum rulelength can be set.
A TwoKey-Plot is a scatterplot of confidence vs. support of a set of rules.
A TwoKey-Plot gives an overview of all the rules.
For example a rule with high confidence and support can be found in the upper right corner.
A DoubleDecker-Plot is a graphical display of a specific rule of the form X1,...Xn -> Y. The display is divided into two parts: Consequence on the top and Antecedent below.
A DoubleDecker-Plot always represents the whole dataset. The specific rule can be found on the right blue barchart. The area of this barchart is equal to the support, the height represents the confidence of the rule.
The other blue barcharts show rules like (not)X1,..(not)Xn -> Y, where not Xi (instead of Xi) is represented by a magenta area in the Antecedent part at the variable Xi, while Xi is represented by a blue area.
All possible combinations of the variables in the Antecedent are shown. Confidence and support can be recognized as described above. The magenta barcharts in the Consequence part show rules of the form (not)X1,..(not)Xn -> not Y.
A list of all selected variables of the dataset and a barchart of their counts is displayed in this graphic.
The ruletable shows information about every actual rule. Barcharts of confidence and support allow to compare different rules easily.
The Detail-Frame gives an overview of the appearence of each item in rules that are shown in the Ruletable. Antecedent and consequence are handled separatly. The Barcharts show the counts.
Different search- and sort-functions are available in PISSARRO.
Pruning is a possibility to reduce the possible large number of rules.
PISSARRO offers different pruning methods, some with statistical background others are easy heuristics.
These methods can be performed by one self or combined to a pruning sequence.
PISSARRO offers a number of interactive tools.
Querying: Get more information in plots and summaries.
Zooming: Zooming is available in some graphical displays (e.g. TwoKey-Plot).
Selection: Selecting rules and variables is possible with different selection modes.
Linking: Plots and summaries a linked.
(C)Copyright 2002 Dept. of Computer Oriented Statistics and Data Analysis, University of Augsburg, All rights reserved.