Skip to main content
def AgentSelectorWorker.transform(self, data) [source]
Generate a fresh selection audit report on any dataset. Re-evaluates the champion agent’s diagnostics (performance, SHAP drift) on a new data sample without re-fitting. Use this to:
  • Monitor the champion on a subsequent OOT window.
  • Detect distribution shift before it impacts production.
  • Generate a fresh audit report for compliance.

​
Parameters

data : pDF
The dataset to evaluate (must contain target + features).
Typically a new OOT sample.

​
Returns

AgentSelectorResult
Updated performance and SHAP diagnostics on the
new data, with the same champion predictor.
No Re-fitting: This method does not re-train the champion. It only re-computes the diagnostics on the new data. The agent_predictor remains the same fitted instance from fit.Use Case: If your OOT window shifts (e.g., next month’s data), call transform on the new sample to verify that the champion’s feature behavior remains stable. If SHAP drift increases, consider re-selecting.