Skip to content
ML Academy

Introduced on day 51

nDCG (normalised discounted cumulative gain)

A rank-sensitive score for graded relevance that discounts gains by position and divides by the best possible ordering, giving a value between 0 and 1.

In detail

DCG@k sums the gain of each result divided by a logarithmic position discount; nDCG@k divides by the DCG of the ideal ordering of all labelled passages. Gains may be linear or exponential in the grade, so two nDCG values are comparable only with the same gain scale, discount and cutoff. It is useful when some passages fully answer the question and others only help.