Implements a simplified Kneedle algorithm to detect the point of maximum
curvature in a 2-D curve. Useful for choosing the optimal number of
top-ranked genes (top-N) in a separation-sweep: the elbow is where
adding more genes yields diminishing returns.
Arguments
- x
Numeric vector of x-values (e.g. N, the number of top genes).
- y
Numeric vector of y-values (e.g. S_AB scores).
Value
An integer index into x and y where the elbow is detected.
Returns 1L if the inputs are too short (< 3 points).
Examples
x <- seq(10, 100, by = 10)
y <- c(0.5, 0.3, 0.15, 0.08, 0.05, 0.03, 0.02, 0.018, 0.015, 0.014)
idx <- find_elbow(x, y)
x[idx] # optimal N
#> [1] 10
