## CN2 discretization question

2 posts
• Page

**1**of**1**### CN2 discretization question

I am currently using the Orange GUI interface, and had a couple questions about the CN2 Algorithm.

1)I currently run the app using WRACC option. From going through the source code and manual I believe this uses CN2SDUnorderedLearnerClass, which is the CN2-SD algorithm developed by Lavrac. Is this correct?

2)Using CN2 with WRACC, is the discretization of continuous attributes conducted by entropy or binarization scheme where every possible attribute value is turned into a boolean feature?

3)If entropy is used, is the entropy-MDL function used in the GUI discretize option?

Thank you again for such a great product.

-Nick

1)I currently run the app using WRACC option. From going through the source code and manual I believe this uses CN2SDUnorderedLearnerClass, which is the CN2-SD algorithm developed by Lavrac. Is this correct?

2)Using CN2 with WRACC, is the discretization of continuous attributes conducted by entropy or binarization scheme where every possible attribute value is turned into a boolean feature?

3)If entropy is used, is the entropy-MDL function used in the GUI discretize option?

Thank you again for such a great product.

-Nick

1)I currently run the app using WRACC option. From going through the source code and manual I believe this uses CN2SDUnorderedLearnerClass, which is the CN2-SD algorithm developed by Lavrac. Is this correct?

Yes, almost as the original CN2-SD algorithm. It uses the weighted relative accuracy formula for rule quality and weighted covering mechanism (weights of covered examples are multiplied). There are some differences, like discretization of attributes - see below.

2)Using CN2 with WRACC, is the discretization of continuous attributes conducted by entropy or binarization scheme where every possible attribute value is turned into a boolean feature?

The discretization is conducted with the entropy-MDL scheme, however not as a preprocessor before learning, but rather each attribute is discretized in each specialization step - so different splits are possible in different rules.

3)If entropy is used, is the entropy-MDL function used in the GUI discretize option?

If you are asking about the "Discretization" widget, then there is the entropy-MDL function available.

Martin

2 posts
• Page

**1**of**1**