ProgramMetricAggregationStrategy
Enums · AiDotNet.Enums
Selects the rule that turns a dictionary of evaluator metrics into one scalar quality.
For Beginners
Your scoring code hands back several measurements, but the search needs a single
number to decide which program is better. This setting says how those measurements are combined. Start with
CombinedScoreOrMean: have your evaluator report a metric called combined_score and that
value is used directly. Use Weighted when you know how much each measurement matters, and
Tchebycheff when you want the search to improve whichever measurement is currently furthest from
your target rather than letting a strong measurement hide a weak one.
How It Works
An evaluator usually reports several numbers - accuracy, latency, memory, a penalty - but a quality-diversity archive ranks candidates on exactly one. This enumeration names the rule used to collapse the dictionary, and naming it is the point: an aggregation that is chosen explicitly can be validated, logged, and reproduced, whereas an aggregation that happens implicitly inside an evaluator cannot.
CombinedScoreOrMean reproduces the reference OpenEvolve rule
(openevolve/utils/metrics_utils.py``get_fitness_score): prefer a metric literally named
combined_score, and otherwise average the numeric metrics that are not archive feature dimensions.
The remaining members are deliberate alternatives with no upstream counterpart.
Fields
| Field | Summary |
|---|---|
ProgramMetricAggregationStrategy CombinedScoreOrMean |
Uses a metric named combined_score when present, and otherwise the mean of numeric metrics. |
ProgramMetricAggregationStrategy Mean |
Always averages the numeric metrics, ignoring any metric named combined_score. |
ProgramMetricAggregationStrategy Tchebycheff |
Scores the largest weighted shortfall from a declared reference point. |
ProgramMetricAggregationStrategy Weighted |
Combines metrics as a weighted mean using explicitly declared, validated weights. |
int value__ |