SCORE/RANK functions – PreferenceSQL

SCORE- and RANK-functions

Creating SCORE- and RANK-functions

To get an overview of SCORE- and RANK-functions see Score and Numerical ranking.

Syntax

Syntax for creating a score function

CREATE [OR REPLACE] SCOREFUNCTION '<name>' AS BEGIN
[IMPORT
<javacode: imports>
#IMPORT]
GETMAXSCORE:
<javacode: getMaximumScore() method>
#GETMAXSCORE
DOMAINTYPE:
<BOOLEAN>| <STRING> | <SET> | <DOUBLE>
#DOMAINTYPE
SCORE:
<javacode: score(Object) method>
#SCORE
END;

Syntax for creating a rank function:

CREATE [OR REPLACE] RANKFUNCTION '<name>' AS BEGIN
[IMPORT
<javacode: imports>
#IMPORT]
RANK:
<javacode: rank method>
#RANK
END;

Example

As a use case we want to represent a customer interested in buying a relative new, inexpensive car. More precisely his favor for a cheap car is three times more important than the age of the vehicle. For the price, a difference of 100 rupees doesn't matter. As for the age, we like to use a regression function like the square root so that newer cars can more likely be differentiated whereas for older cars the age difference won't really matter.

To represent this preference, we can use a rank function with the predefined sqrtScoreF scoring function for the age:

carRankF : t ↦ (3⋅priceScoreF(t[price]) + sqrtScoreF(t[quality]))/4

where t ∈ Price × Quality

Thus we can write:

CREATE RANKFUNCTION 'carRankF' AS BEGIN
IMPORT:
import java.util.Iterator;
import java.util.List;
#IMPORT
RANK:
Iterator<RankFunctionParameter> par = getParameters().iterator();
double price = par.next().getFunctionValue(objPref, prefContext);
double age = par.next().getFunctionValue(objPref, prefContext);
return (price*3+age)/4;
#RANK
END;

It remains to define the scoring functions for the price:

CREATE SCOREFUNCTION 'priceScoreF' AS BEGIN
IMPORT:
import preference.dataconnector.MinMax;
#IMPORT
GETMAXSCORE:
MinMax minmax = getDataConnector().getMinMaxValues(getTableName(), getAttributeName());
return minmax.getMax()/100.;
#GETMAXSCORE
DOMAINTYPE:
double
#DOMAINTYPE
SCORE:
double value = Double.parseDouble(A.toString()); // we have A : Object and cast it to double
return value/100.;
#SCORE
END;

Finally, we want to search with our new rank function by:

SELECT *
FROM car
PREFERRING (price SCORE 'priceScoreF' | age SCORE 'sqrtScoreF') RANK 'carRankF';

Predefined SCORE-functions

absScoreF: Returns absolute values
identityScoreF: Returns unmodified score-values
negScoreF: Negates the values
powThreeScoreF: Values to the power of 3
powTwoScoreF: Values to the power of 2
sqrtScoreF: Square root of the values

Predefined RANK-functions

sumRankF: Does a summation or a weighted summation, depending on the syntax; for a standard summation just use it as any other rank-function, for weighted summation use it as with the following addendum:
<number>[,<number>]∗

The number of attributes (here: 2, age and mileage) have to match the number of arguments (here: 2; 2, and 3).

SELECT *
FROM CAR
PREFERRING (age SCORE 'identityScoreF' | mileage SCORE 'identityScoreF') RANK 'sumRankF' '2,3';

avgRankF: Returns the average or the weighted average, for latter see syntax for "sumRankF"
deviationRankF: Calculates the deviation
maxRankF: Returns the highest value
minRankF: Returns the lowest value
productRankF: Multiplies all values and returns the arithmetic product
varRankF: Calculates the variance

Function Management

The following command shows all selfdefined (e.g. dynamic) score and rank functions: