I hope you have tested the current (rudimentary) version of Numerical Cruncher and have lots of suggestions to do.
I would like to begin design work in order to create a new base upon which we can develop new algorithms and techniques, as well as bridges and adapters to existing systems such as Weka.
Regarding to the user interface, I suggest we could develop a web-based system (since it is easier to manage than a windows-based interface).
Things to be done:
- Dataset modelling (I suggest a model which will be published in a forthcoming issue of Communications of the ACM). You can take a look at it: http://elvex.ugr.es/etexts/English/olap-oltp.pdf
- Component-based kernel
- Key classes standardization: Classifier, Clusterer...
- Bridges to existing systems (e.g. Weka)
- Easy-to-use interface, with a back-end recommender system (if possible) which should guide users through dataset exploration.