Anytime Nearest Neighbor Classifier

Author: Reuben Bell, Liang Liu

This program is the result of our term project of comp523 data stream mining at the university of Waikato 2012. And can be used as an extension of MOA(Massive Online Analysis).

This program implemented anytime algorithm described in the following paper:

Shieh, J. and Keogh, E. (2010). Polishing the Right Apple: Anytime Classification Also Benefits Data Streams with Constant Arrival Times. In Proceedings of the 2010 IEEE International Confer-ence on Data Mining (ICDM '10). pp. 461-470. Washington, DC, USA. IEEE Computer Society.

The structure of our program:

MOA
    | ---- PartialResult
    | ---- classifiers ----- anytime --- AbstractAnytime
    | ---- NearestNeighbor
    | ---- tasks       ----- EvaluatorAnytime

We implemented the data structure described in the paper cited above in a separate class PartialResult. We chose to create a task class EvaluatorAnytime to implement the selection scheme. For more detail see either origin paper cited above or our report: comp523_Report.pdf.

Test result:

We found this program could work with most stream generator in MOA, but not all the stream generator and not all the stream generator settings.

This is our code: code.zip.

And how to use it: install guide.pdf.

Contact

Reuben Bell reuben_bell at hotmail dot com
Liang Liu     liangliu230 at gmail dot com