Weighted Average

Computes a weighted average over a categorical label field. Each label is mapped to a numeric weight, and the result is the mean of those weights across all samples. Raises an error if an unknown label is encountered.

Output

A single metric named {field}_weighted_average by default, or the value of name if provided.

Examples

Example: Data Leakage Detection. A Python scorer classifies each sample as "No leakage", "Partial leakage", or "Full Leakage". The weighted average maps these labels to numeric scores so a single metric captures overall leakage severity.

scorers:
  - type: python
    compute_scores_snippet: |
      def compute_scores(sample, solver_output):
          # ... classification logic ...
          return {"label": "No leakage"}  # or "Partial leakage" / "Full Leakage"
    metrics:
      - type: weighted_average
        field: label
        name: "Data Leakage Safety"
        weights:
          "No leakage": 1.0
          "Partial leakage": 0.5
          "Full Leakage": 0.0

Configuration

Properties

type Literal "weighted_average" required

The type of the metric.

field string, TemplateValue required

The field containing the labels over which to compute the weighted average.

weights object, TemplateValue required

A mapping from label to numeric weight. The weighted average is computed as sum(weight_i) / count.

name string, TemplateValue

The name given to the metric value. If not specified, it is {field}_weighted_average.

key string

Unique identifier assigned to the entity in AI GO!.