Comparative Metric
Jump to navigation
Jump to search
A Comparative Metric is an evaluation metric that assesses performance through relative comparison between multiple outputs, systems, or models.
- AKA: Relative Metric, Comparison Metric, Relational Measure.
- Context:
- It can typically facilitate Model Selection through pairwise comparisons and ranking generation.
- It can typically identify Relative Advantages via preference scoring and superiority assessment.
- It can typically support Tournament Evaluations through head-to-head comparisons and league tables.
- It can often enable Preference Learning via comparison feedback and ranking signals.
- It can often provide Directional Guidance through improvement indicators and regression detection.
- ...
- It can range from being a Binary Comparative Metric to being a Multi-Way Comparative Metric, depending on its comparison cardinality.
- It can range from being a Ordinal Comparative Metric to being a Cardinal Comparative Metric, depending on its comparison value type.
- It can range from being a Symmetric Comparative Metric to being an Asymmetric Comparative Metric, depending on its comparison directionality.
- It can range from being a Local Comparative Metric to being a Global Comparative Metric, depending on its comparison scope.
- ...
- Example(s):
- Counter-Example(s):
- Absolute Metric, which measures against fixed standards rather than other outputs.
- Threshold Metric, which uses fixed cutoffs rather than relative comparison.
- Ground Truth Metric, which compares to ideal rather than to alternatives.
- See: Evaluation Metric, Relative Evaluation Method, Pairwise Learning-to-Rank Algorithm, Ranking Task, Preference Learning, Tournament Selection, Performance Metric.