AHR: Calculate AHR (or expected AHR)

In EPX: Ensemble of Phalanxes

Description

Calculates average hitrate (AHR), which is equivalent to average precision. When there are ties in ranking (not all values in phat are unique), the result is the expection of AHR. The algorithm that produces this analytic result assumes that the items in any tied group are in an arbitrary order within the group.

Usage

AHR(y, phat, ...)

Arguments

y	True (binary) response vector where 1 is the rare/relevant class.
phat	Numeric vector of estimated probabilities of relevance.
...	Further arguments passed to or from other methods.

Details

Implementation adapted from Wang (2005, Chapter 3). Please also see Chapter 3 of Wang (2005) for AHR and expected AHR formulas.

Value

Numeric value of average hitrate; expected average hitrate when there are ties.

References

Wang, M. (2005). Statistical Methods for High Throughput Screening Drug Discovery Data (Doctoral thesis). University of Waterloo, Waterloo, Ontario, Canada.

Examples

## AHR when there are no ties in phat:
resp <- c(1, 0, 0, 0, 1)
prob <- (1:5)*0.1
AHR(y = resp, phat = prob)
# expect answer: 1/2 * (1 + 0 + 0 + 0 + 2/5)

## (Expected) AHR when there are ties in phat:
resp <- c(1, 1, 0,   0,   0,   0,   0,    1,   0, 0)
prob <- c(1, 1, 1, 0.4, 0.4, 0.3, 0.2, 0.15, 0.1, 0)
AHR(y = resp, phat = prob)
# expect answer: 1/3 * (2/3 + 1/2 * (1/3 + 2/3) + 1/3 * 4/3 +
#                       1/8 * (2/3 + 2/3 + 2/3 + 1))

EPX documentation built on July 7, 2021, 1:06 a.m.