gissimo / nsc

Python implementation of the Nearest Subclass Classifier (NSC), based on Maximum Variance Clusters (MVC).

Nearest Subclass Classifier

Python implementation of the Nearest Subclass Classifier (NSC), based on Maximum Variance Clusters (MVC).

The classification algorithm is based on the work published in the papers:

• The nearest subclass classifier: a compromise between the nearest mean and nearest neighbor classifier (Veenman and Reinders, 2005).

• A maximum variance cluster algorithm (Veenman, Reinders, Backer, 2002).

Follows a brief description of how the scripts work, however probably nothing of this is going to help you much unless you are already familiar with the algorithm.

• do_mvc.py creates the prototypes from a text file. Requires as inputs:
  1. filename
  2. value of sigma-square-max
  3. column separator
  4. position of class identifier within the line
• do_nsc.py classifies points from a file, given a file with prototypes. Requires as inputs:
  1. file of non-classified points
  2. prototypes file
  3. column separator
• do_cross.py finds the optimal value of sigma-square-max. Requires as inputs:
  1. file with points
  2. number of iterations
  3. column separator
  4. position of class identifier within the line
• do_test.py tests NSC/MVC on a set of already-classified points. Requires as inputs:
  1. file with points
  2. value of sigma-square-max
  3. column separator
  4. position of class identifier within the line

More details (in Italian) here.