igd: Inverted Generational Distance (IGD and IGD+) and Averaged...

Description Usage Arguments Details Value Author(s) References Examples

View source: R/igd.R

Description

Functions to compute the inverted generational distance (IGD and IGD+) and the averaged Hausdorff distance between nondominated sets of points.

Usage

1
2
3
4
5
igd(data, reference, maximise = FALSE)

igd_plus(data, reference, maximise = FALSE)

avg_hausdorff_dist(data, reference, maximise = FALSE, p = 1L)

Arguments

data

(matrix | data.frame)
Matrix or data frame of numerical values, where each row gives the coordinates of a point.

reference

(matrix | data.frame)
Reference set as a matrix or data.frame of numerical values.

maximise

(logical() | logical(1))
Whether the objectives must be maximised instead of minimised. Either a single logical value that applies to all objectives or a vector of logical values, with one value per objective.

p

(integer(1)) Hausdorff distance parameter (default: 1L).

Details

The generational distance (GD) of a set A is defined as the distance between each point a \in A and the closest point r in a reference set R, averaged over the size of A. Formally,

GD(A,R) = (1/|A|) * ( sum_{a in A} min_{r in R} d(a,r)^p )^(1/p)

where the distance in our implementation is the Euclidean distance:

d(a,r) = sqrt( sum_{k=1}^M (a_k - r_k)^2)

The inverted generational distance (IGD) is calculated as IGD_p(A,R) = GD_p(R,A).

The modified inverted generational distanced (IGD+) was proposed by \citetIshMasTanNoj2015igd to ensure that IGD+ is weakly Pareto compliant, similarly to epsilon_additive() or epsilon_mult(). It modifies the distance measure as:

d^+(r,a) = sqrt(sum_{k=1}^M (max {r_k - a_k, 0 })^2)

The average Hausdorff distance (Δ_p) was proposed by \citetSchEsqLarCoe2012tec and it is calculated as:

Δ_p(A,R) = \max\{ IGD_p(A,R), IGD_p(R,A) \}

IGDX \citepZhoZhaJin2009igdx is the application of IGD to decision vectors instead of objective vectors to measure closeness and diversity in decision space. One can use the functions igd() or igd_plus() (recommended) directly, just passing the decision vectors as data.

There are different formulations of the GD and IGD metrics in the literature that differ on the value of p, on the distance metric used and on whether the term |A|^{-1} is inside (as above) or outside the exponent 1/p. GD was first proposed by \citetVelLam1998gp with p=2 and the term |A|^{-1} outside the exponent. IGD seems to have been mentioned first by \citetCoeSie2004igd, however, some people also used the name D-metric for the same concept with p=1 and later papers have often used IGD/GD with p=1. \citetSchEsqLarCoe2012tec proposed to place the term |A|^{-1} inside the exponent, as in the formulation shown above. This has a significant effect for GD and less so for IGD given a constant reference set. IGD+ also follows this formulation. We refer to \citetIshMasTanNoj2015igd and \citetBezLopStu2017emo for a more detailed historical perspective and a comparison of the various variants.

Following \citetIshMasTanNoj2015igd, we always use p=1 in our implementation of IGD and IGD+ because (1) it is the setting most used in recent works; (2) it makes irrelevant whether the term |A|^{-1} is inside or outside the exponent 1/p; and (3) the meaning of IGD becomes the average Euclidean distance from each reference point to its nearest objective vector). It is also slightly faster to compute.

GD should never be used directly to compare the quality of approximations to a Pareto front, as it often contradicts Pareto optimality. We recommend IGD+ instead of IGD, since the latter contradicts Pareto optimality in some cases (see examples below), but we implement IGD here because it is still popular due to historical reasons. We are not aware of any proof of whether Δ_p(A,R) contradicts or not Pareto optimality, thus it must be used with care.

Value

(numeric(1)) A single numerical value.

Author(s)

Manuel López-Ibáñez

References

\insertAllCited

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
# Example 4 from Ishibuchi et al. (2015)
ref <- matrix(c(10,0,6,1,2,2,1,6,0,10), ncol=2, byrow=TRUE)
A <- matrix(c(4,2,3,3,2,4), ncol=2, byrow=TRUE)
B <- matrix(c(8,2,4,4,2,8), ncol=2, byrow=TRUE)
plot(ref, xlab=expression(f[1]), ylab=expression(f[2]),
     panel.first=grid(nx=NULL), pch=23, bg="gray", cex=1.5)
points(A, pch=1, cex=1.5)
points(B, pch=19, cex=1.5)
legend("topright", legend=c("Reference", "A", "B"), pch=c(23,1,19),
       pt.bg="gray", bg="white", bty = "n", pt.cex=1.5, cex=1.2)
cat("A is better than B in terms of Pareto optimality,\n however, IGD(A)=",
    igd(A, ref), "> IGD(B)=", igd(B, ref),
    ", which contradicts it.\nBy contrast, IGD+(A)=",
    igd_plus(A, ref), "< IGD+(B)=", igd_plus(B, ref), ", which is correct.\n")

# A less trivial example.
extdata_path <- system.file(package="eaf","extdata")
path.A1 <- file.path(extdata_path, "ALG_1_dat.xz")
path.A2 <- file.path(extdata_path, "ALG_2_dat.xz")
A1 <- read_datasets(path.A1)[,1:2]
A2 <- read_datasets(path.A2)[,1:2]
ref <- filter_dominated(rbind(A1, A2))
igd(A1, ref)
igd(A2, ref)

# IGD+ (Pareto compliant)
igd_plus(A1, ref)
igd_plus(A2, ref)

# Average Haussdorff distance
avg_hausdorff_dist(A1, ref)
avg_hausdorff_dist(A2, ref)

eaf documentation built on May 7, 2021, 5:06 p.m.