Feature roles
Understand how Ounias interprets dataset columns and why their assigned roles affect downstream analyses.
Why feature roles matter
Columns in a structured dataset can serve different analytical purposes.
During ingestion, Ounias detects likely column types and allows you to review how each column should participate in analysis.
The selected roles determine which columns are available to analytical pipelines and which have special meanings such as identifiers, supervised targets, or longitudinal ordering variables.
Analytical features
These columns provide information to analytical methods.
- Continuous features
- Categorical features
- Binary features
- Note: Ordinal features are not explicitly supported. Numerically encoded ordinal variables can be assigned as continuous, preserving their ordering while also treating numerical spacing as meaningful.
Special roles
Some columns describe the structure of the analysis rather than acting as analytical features.
- Identifier/ID — identifies each observation in a standard dataset, or the repeated entity in a longitudinal dataset
- Supervised target — classification/regression
- Observation index — identifies the sequence of observations within an entity
- Observation datetime — provides the timestamp used to order observations and preserve temporal spacing
Excluded columns
Not every column needs to participate in analysis.
Columns that are irrelevant, administrative, duplicated, unsuitable for modeling, or otherwise outside the intended analysis can be excluded during feature mapping.
Validation
Ounias checks whether configured roles are consistent with the underlying data.
Feature-role validation is intended to catch mismatches before they reach analytical pipelines. Incorrect feature roles can materially change which methods are available and how results should be interpreted.