Name | Object | Symbol |
---|---|---|

Chebyshev Distance | `org.apache.mahout.math.algorithms.common.distance.Chebyshev` |
`'Chebyshev` |

Cosine Similarity | `org.apache.mahout.math.algorithms.common.distance.Cosine` |
`'Cosine` |

In Mahout one can access the distant metrics directly to measure the distance between two arbitrary vectors, or
can specify which distance metric to use as part of an algorithm. In the latter case the distance metric is called
by `Symbol`

, we never pass Distance metrics directly to an algorithm. This design choice, in part has to do with
serialization of object and keeping the engine bindings as simple as possible. Behind the scenes, the only thing
that is serialized and sent to the workers is a number which specifies what distant metric to use- this is much more
abstract and easier to maintain on the back end than making sure each function can be serialized by any arbitrary engine.
We feel from the user perspective, it may seem quirky but causes no decrease in usability. If a user wishes to use a
custom distance metric- simply add it to math-scala/src/main/org/apache/mahout/math/common/DistanceMetrics.scala
and recompile.

**Meausring the distance between two vectors**

```
import org.apache.mahout.math.algorithms.common.distance._
val v1 = dvec(1.0, 1.5, -1.2, 3.5)
val v2 = dvec(0.1, -1.4, 10.5, 3.2)
Cosine.distance(v1, v2)
```

**Using distance in clustering**

```
import org.apache.mahout.math.algorithms.clustering.CanopyClustering
val drmA = drmParallelize(dense((1.0, 1.2, 1.3, 1.4),
(1.1, 1.5, 2.5, 1.0),
(6.0, 5.2, -5.2, 5.3),
(7.0,6.0, 5.0, 5.0),
(10.0, 1.0, 20.0, -10.0)))
val model = new CanopyClustering().fit(drmA, 'distanceMeasure -> 'Cosine)
```