ML Wiki
Machine Learning Wiki - A collection of ML concepts, algorithms, and resources.

Hurwitz's Index

Hurwitz’s Index

How to choose an alternative?

Idea: in-between the extreme pessimism and extreme optimism

Coefficient of Pessimism:

  • define $\alpha \in [0, 1]$ as the coefficient of pessimism
  • then choose the solution $a \in A$ that is the closest to
  • $\max_{a \in A} \big[ \alpha \cdot \min_{e \in E} c(a, e) + (1 - \alpha) \cdot \max_{e \in E} c (a, e) \big]$

Example:

$\alpha = 0.5$

$c$ $e_1$ $e_2$ $e_3$ $\alpha =0.5$   $a_1$ 40 70 -20 (-20 + 70) / 2 25   $a_2$ -10 40 100 (-10 + 100) / 2 45   $a_3$ 20 40 -5 (-5 + 40) / 2 17.5         max: 45 $\to a_2$

Downsides

  • bad use of information
  • now the scale matters - since we multiply by $\alpha$
  • how to determine and justify $\alpha$?

Sources