AgentSkillsCN

ml-alpha

适用于在流动性充裕的市场中，采用有监督与集成信号模型，并结合泄漏安全的训练与执行感知评估的ML阿尔法工作流。当任务涉及基于机器学习的阿尔法生成与验证时，可选用此类工作流。

SKILL.md
--- frontmatter
name: ml-alpha
description: "ML alpha workflows for supervised and ensemble signal models in liquid markets with leakage-safe training and execution-aware evaluation. use when tasks involve machine-learning based alpha generation and validation."

ML Alpha

objective

Develop machine-learning alpha models with robust validation, calibration, and deployment controls.

workflow

  1. •define target construction and purged train-validation splits.
  2. •engineer predictive features with leakage and survivorship safeguards.
  3. •train and calibrate models with benchmark challengers.
  4. •evaluate alpha net of costs under realistic execution assumptions.
  5. •deploy only when live monitoring and drift controls are configured.

required diagnostics

  • •precision, calibration, and information-coefficient stability.
  • •feature drift and target drift diagnostics.
  • •out-of-sample decay and horizon-specific performance.
  • •cost-adjusted edge versus simple baseline models.
  • •live-versus-backtest discrepancy monitoring.

risk controls

  • •enforce purged validation and embargo rules.
  • •enforce model registry and reproducible training artifacts.
  • •enforce rollback trigger on live calibration breakdown.

outputs

  • •run python scripts/ml_alpha_diagnostics.py input.csv --output diagnostics.json and keep the json artifact.
  • •write an implementation memo using references/ml-alpha-playbook.md with assumptions, tests, limits, and rollout plan.

resources

  • •use scripts/ml_alpha_diagnostics.py for deterministic diagnostics.
  • •use references/ml-alpha-playbook.md for the domain checklist and delivery structure.