Cross-method consensus
Understand how Ounias compares labeling results and constructs consolidated meta-labels.
Purpose and inputs
Consensus works with a selected set of analytical runs and investigates how their labeling outputs relate to one another.
It examines recurring co-grouping patterns and disagreement between labelings rather than treating any individual run as definitive.
Consensus modes
Mode selection controls which labeling results can participate in the consensus analysis.
- Fixed mode — select runs under a specified number of labels.
- Free mode — select runs from methods whose number of labels is not predetermined.
Analysis and outputs
The multi-phase consensus analysis examines relationships between labels across selected runs, including the fraction of observations grouped together and evidence concerning the number of labels.
Ounias constructs a meta-label artifact from the selected results and creates an associated meta-run. This result can be inspected alongside its contributing runs.
How to interpret consensus
Agreement between methods is evidence of consistency across selected analytical approaches, not proof that a grouping represents a real-world category.
The interpretation depends on the methods and configurations included, their shared assumptions, and properties of the dataset.