-
-

Studying patterns

+
+

Bootstrapping and rule robustness

+

There are two sources of uncertainty that we are interested to quantify when using an ex-Fuzzy classifier if we want reliable predictions. This library offers tools to:

+
    +

  1. Quantify the uncertainty due to the variability in the rulebases generated by the genetic optimization process.

    2. +

  2. Quantify the uncertainty due because of variability in the data.

    3. +
+

The first type of uncertainty can be analyzed using the pattern stability tools described below. The second type of uncertainty can be analyzed using Bootstrapping methods.

+
+

Variability in the rulebases due to the genetic optimization

Due to the nature of the genetic optimization, we might obtain different rulebases for the same dataset if we run the experiment with different seeds. Due to the nature of rule-based reasoning, we can directly study how often some patterns appear and how useful variables are by directly analyzing the different rulebases and the way in which their rules use each variable.

The module ex_fuzzy.pattern_stability contains a series of functions to analyze the ocurrences of the patterns when the same experiment is run with different seeds. This module comes with the class ex_fuzzy.pattern_stability.pattern_stabilizer whose initialization sets the parameters for the analysis. Then, the function ex_fuzzy.pattern_stability.stability_report can be used to run the experiments. Finally, use ex_fuzzy.pattern_stability.pattern_stabilizer.text_report to obtain a text report of the rules obtained and how often they appeared in the experiments. Use ex_fuzzy.pattern_stability.pattern_stabilizer.pie_chart_class and ex_fuzzy.pattern_stability.pattern_stabilizer.pie_chart_var to obtain pie charts of the ocurrences of the variables per class according to the linguistic variables used.

This is an example of a textual report:

@@ -139,10 +153,20 @@