CobWeb : Access to node elements in the tree

[kumar.saurabh]

Hi,

While using Cobweb, I am not able to get access to tree nodes by obtaining the tree root.
I can get the entire graph as a string but it is not an efficient method as i need to parse through it and then repeat calculations.
Is there a method to obtain the standard deviations at each node to be able to derive the "concepts" at different tree levels.

Thanks,
Saurabh

[Mark]

You will need to hack the code I'm afraid. A bunch of member variables in the inner CNode class will need accessor methods or made public.

Cheers,
Mark.

[kumar.saurabh]

Hi,

Is there a standard dataset against which the CLASSIT implementation has been validated (for some acuity value)?
In Gennari's paper, it is mentioned that the algorithm is designed for numeric values and does not handle nominal values.
So how does the Weka implementation handle the combination of Numeric and Nominal values (what is the category utility function)?
The results of clustering on the same dataset after using a filter to convert Numeric values (0/1) to nominal are clearly different.

[Mark]

You need to take a look at the Cobweb paper as well:

D. Fisher (1987). Knowledge acquisition via incremental conceptual clustering. Machine Learning. 2(2):139-172.

If you look around, you'll find that there aren't many implementations of Cobweb/Classit out there :-) Perhaps not surprising given that there are a number of differences and ambiguities between the two papers. The assumptions made by the Weka implementation are in the javadoc and the help info in the GUI.

Cheers,
Mark.

[kumar.saurabh]

Thanks Mark.
Are there alternative algorithm for conceptual clustering?
Is it possible to use the Hierarchical Clusterer in Weka without specifying the exact number of clusters? How can i retrieve the hierarchy programmatically using the API?