Class Ranking

java.lang.Object

com.eqixiac.equinix.design.solve.Ranking

public final class Ranking
extends Object

The scoring rules shared by ExhaustiveSolver and SensitivityAnalysis, so a re-ranking under other weights or over a relaxed set uses exactly the arithmetic the solver used.

• Normalisation: per objective, 0..100 (100 = best) on the objective's own ValueScale: a fixed scale where the objective declares one (a percentage, a ratio to the best cost, an anchored latency), otherwise the spread of the set being ranked; an unknown value has no score (NaN).
• Weighted score: sum(weight x score) / sum(weight) over the objectives with a positive weight; an unknown score contributes nothing and its weight is not redistributed, so missing data never raises a score.
• Order: higher weighted score; then Pareto-optimal before dominated (so the recommendation is never a dominated option that merely ties); then fewer metros; then portfolio key.

Method Summary

Modifier and Type	Method	Description
static int	best(List<double[]> rows, Direction[] directions, double[] weights, List<MetroPortfolio> portfolios)
static int	best(List<double[]> rows, Direction[] directions, ValueScale[] scales, double[] weights, List<MetroPortfolio> portfolios)
static double[][]	normalise(List<double[]> rows, Direction[] directions)
static double[][]	normalise(List<double[]> rows, Direction[] directions, ValueScale[] scales)
static Comparator<Integer>	order(double[] weighted, boolean[] pareto, List<MetroPortfolio> portfolios)
static boolean[]	paretoFlags(List<double[]> rows, Direction[] directions)
static double[]	weighted(double[][] scores, double[] weights)

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

normalise

public static double[][] normalise(List<double[]> rows,
 Direction[] directions)

Parameters:

rows - raw values, one row per option, NaN = unknown

directions - per objective

Returns:

normalised scores, same shape, NaN where the value is unknown

normalise

public static double[][] normalise(List<double[]> rows,
 Direction[] directions,
 ValueScale[] scales)

Parameters:

rows - raw values, one row per option, NaN = unknown

directions - per objective

scales - per objective; null (or a null entry) means the spread of the set

Returns:

normalised scores, same shape, NaN where the value is unknown

weighted

public static double[] weighted(double[][] scores,
 double[] weights)

Parameters:

scores - normalised scores

weights - per objective, non-negative

Returns:

the weighted score of each option (0 for all when every weight is 0)

order

public static Comparator<Integer> order(double[] weighted,
 boolean[] pareto,
 List<MetroPortfolio> portfolios)

Parameters:

weighted - weighted scores

pareto - whether each option is Pareto-optimal in the set

portfolios - the options' portfolios

Returns:

the ranking order as a comparator over option indexes, best first

best

public static int best(List<double[]> rows,
 Direction[] directions,
 double[] weights,
 List<MetroPortfolio> portfolios)

Parameters:

rows - raw values

directions - per objective

weights - per objective

portfolios - the options

Returns:

the index of the option the rules recommend, or -1 for an empty set

best

public static int best(List<double[]> rows,
 Direction[] directions,
 ValueScale[] scales,
 double[] weights,
 List<MetroPortfolio> portfolios)

Parameters:

rows - raw values

directions - per objective

scales - per objective; null means the spread of the set

weights - per objective

portfolios - the options

Returns:

the index of the option the rules recommend, or -1 for an empty set

paretoFlags

public static boolean[] paretoFlags(List<double[]> rows,
 Direction[] directions)

Parameters:

rows - raw values

directions - per objective

Returns:

for each row, whether no other row dominates it