Building Composite Objective Scores for Hyperparameter Search

Summary

This brief Q&A explains how to change the scoring function used during hyperparameter search. The question asks how to include annualized return, win rate, and maximum drawdown alongside a Sharpe-based score. The response says the score function can read available performance fields from the search result and return a chosen metric, or combine multiple metrics into a new score.

The examples focus on retrieving maximum drawdown or multiple performance fields, but the note does not specify how to scale, weight, or combine metrics with different units and objectives. It also does not discuss whether drawdown should be minimized, how to handle missing fields, or how to prevent selection bias during repeated tuning. The guidance is therefore about configuring the objective interface rather than prescribing a robust scoring methodology.

Key ideas

• A hyperparameter search score can be customized to use performance metrics present in the results.
• The example selects maximum drawdown as the score.
• Multiple performance metrics can be retrieved or combined into a composite objective.
• The note does not explain metric scaling, weighting, or directionality in a composite score.
• Metrics must be available in the search result before they can be used.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.