Grouping Stock Observations by Date for Learning-to-Rank Models
Summary
The post asks how to define groups for a learning-to-rank model in a stock selection setting. In ranking algorithms, each group contains observations that should be ranked against one another; the author proposes grouping stock samples by trading date so that stocks from the same date form a cross-sectional ranking set.
The question focuses on BigQuant's stockRanker interface, where the visible settings appear to expose ordinary boosting-tree parameters but not an explicit group field. The author also asks whether validation data needs its own group definitions. The post does not include an answer, implementation details, or empirical results, so it identifies a modeling and interface question rather than resolving it. Its useful takeaway is the distinction between defining ranking groups and tuning the underlying tree model, with date-based grouping presented as the author's intended design for stock ranking.
Key ideas
- Learning-to-rank models compare observations within defined groups.
- The post proposes grouping stock observations by trading date for cross-sectional ranking.
- It questions where group definitions are configured in BigQuant stockRanker.
- It also asks whether validation data requires separate group assignments.
- The post provides no answer or empirical evidence.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.