[source]

Canberra

A weighted version of the Manhattan distance, Canberra examines the sum of a series of fractional differences between two samples. Canberra can be very sensitive when both coordinates are near zero.

\[ Canberra(\mathbf {a} ,\mathbf {b} )=\sum _{i=1}^{n}{\frac {|a_{i}-b_{i}|}{|a_{i}|+|b_{i}|}} \]

Data Type Compatibility: Continuous

Parameters

This kernel does not have any parameters.

Example

use Rubix\ML\Kernels\Distance\Canberra;

$kernel = new Canberra();

References

---

1. G. N. Lance et al. (1967). Mixed-data classificatory programs I. Agglomerative Systems. ↩