skala

Simple Extensions for Scala

Build

We are using SBT to build the project. Although, you can build it just using Scalac.

SBT

Simple Build Tool is the most popular build tool for scala. We respect that. So for compiling the code:

$ sbt compile

And for testing:

$ sbt test:run

SBT respects our own SUnit tests. For making a jar out of the project, do:

$ sbt package

Current Things

Currently we have implemented two thing:

PairedIterable

Actually, ir.angellandros.scala.collection.PairedIterable. The main reason for such a data structure is to have reduceByKey. You can now do this:

import ir.angellandros.scala.collection.Implicits._
val l = List(1,1,1,2,2,3)
l.map(_ -> 1).reduceByKey(_+_)

Keyed Vector

The idea behind KeyedVector is that getting rid of dictionary. You don't need a dictionary, because the indices could be String or any other thing that you want. Although it could be Int and you can use it with some dictionary. Anyway, KeyedVector has a close relation with HashMap. Every KeyedVector has an ID, and this is the most important difference:

val vector = new KeyedVector(id, map)

and with vector.id or vector.toMap you can get either one of them. You can also do vector.get(key) or vector.keySet.

The other class implemented here is KeyedVectors, that provides some tools for KeyedVectors:

val i1 = KeyedVectors.dot(v1, v2)
val i2 = KeyedVectors.normInfinity(v1)
val i3 = KeyedVectors.normNN(v1, n) 	// Minkowski norm ^n, which is squared Euclidean norm for n=2
val i4 = KeyedVectors.distNN(v1, v2, n) // distance function induces from norm
val i5 = KeyedVectors.norm(v1) 			// Euclidean norm
val i6 = KeyedVectors.eucDist(v1, v2)
val i7 = KeyedVectors.sqEucDist(v1, v2) // squared Euclidean distance
val i8 = KeyedVectors.cosineSim(v1, v2)
val i9 = KeyedVectors.cosineDist(v1, v2)

Canopy Clustering Algorithm

Canopy algorithm on pure scala. There are two implementations for canopy algorithm. First, the simple implementation with O(n^2), and the other, the merged implementation with O(n^{3/2}) running time.

val canopy = new CanopyDriver(0.95, 0.5)
val results = canopy.mergedRun(vectors, KeyedVectors.cosineDist[String])

As you can see, the distance function is an input argument.

SUnit

Simple unit testing class. Now has assertEquals.