Building a Recommendation Engine: The XELOPES Library - Realtime Data Mining - page 242

Warehouse Process Warehouse Operation

Management

Transformation OLAP Data

Mining

Business
Nomenclature

Information
Visualization

Analysis

Relational Record Multidimensional XML

Resource

Business
Information

Data
Types

Software
Deployment

Key
Indexes

Expres
sions

Type
Mapping

Foundation

Core Behavioral Relationships Instance

Object

Fig. 12.6 CWM packages used in MiningVector

Example 12.5 Example of a mining vector for the meal basis of the previous
section:

// Create and fill value vector:
double[] mealValues ¼ new double[3];
mealValues[0] ¼ 33000; // calory number
mealValues[1] ¼ 5; // 5 guests
mealValues[2] ¼ cutlery.getKey( new Category("spoon") );
// spoon

// Create mining vector object with values:
MiningVector mealVector ¼ new MiningVector( mealValues );

// Add 'meal' metadata to mining vector:
mealVector.setMetaData( meal );

// Show (double) values of mining vector:
for (int i ¼ 0; i < mealVector.getValues().length; i++)
System.out.println("value["+i+"]

¼

" + mealVector.

getValue(i)); ■

For sparse vectors, i.e., vectors which mainly contain zero coordinate values, the
class MiningSparseVector could be used which extends MiningVector . It stores
sparse vectors more efficiently by means of an additional array of indexes of the
nonzero coordinate values. For binary sparse vectors, i.e., sparse vectors where the
nonzero values are always one, the class MiningBinarySparseVector should be
utilized which in turn extends MiningSparseVector .

12.1.2.3 The Data Matrix: MiningInputStream

So far we have defined the class MiningVector that models a data vector. In order to
model a whole data matrix, we use the abstract MiningInputStream class. MiningIn-
putStream is a virtual collection of mining vectors. Like each of its mining vectors,
MiningInputStream contains a reference metaData to the basis of the attribute space.