In RNG (Relative Neighbourhood Graph) two points, xi
and xj, are neighbours if and only if there is
no other point from set laying in a lune between xi,
xj which is defined as the disjoint intersection
between two hyperspheres centered at xi and xj
and whose radius are equal to the distance between them.
Let q a new pattern to classify, then a classifier based on RNG can
be expressed as follows:
- Search for all the neighbours of q.
- Assign q to the class with a majority of votes from each graph
neighbours.
How to use this applet?
- The top panel is Modes panel, the right panel is Controls
panel and the central panel is Drawing panel.
- To select the number of classes (label No classes) in the controls
panel for the training set.
- To do click over any radiobutton that identify a class (labels 1, 2,
..., 12) in the controls panel.
- To do click over the drawing panel to insert prototypes in the training
set (of the class selected in the step 4). These prototypes will have 2
characteristics whose values will remain defined for the axes x and
y.
- After that you has selected the training set (for all classes) to do click
over the radiobutton sample (label sample) in the controls panel.
- To do click over the drawing panel to insert the sample. Then will appear
the sample's neighbours and the sample classificated.
- The radiobutton with label Inspection in the controls panel allows
to know the coordinates of the next prototype (o sample) to the cursor mouse.
- The radiobutton with label Remove in the controls panel allows
delete the next prototype (o sample) to the cursor mouse.
- The button Clear in the controls panel allows delete all the training
set.