标签设计

Lesson Overview

In "Machine Learning for Trading" → "特征工程", this lesson "标签设计" builds knowledge you can apply immediately.

标签可用未来信息定义结果，但样本划分必须考虑标签终点。未来5步收益标签会跨时间边界，按特征行简单切分仍可能泄漏。 明确预测起点t与终点t+5→只让训练标签终点早于测试起点→冻结阈值→记录末训练和首测试时点。 本课作业：测试特征从168开始，标签跨度5，训练标签终点必须<168。训练特征最晚可为多少？

Core Concepts

These points align with the video walkthrough—take notes as you watch.

• ▸标签可用未来信息定义结果，但样本划分必须考虑标签终点。未来5步收益标签会跨时间边界，按特征行简单切分仍可能泄漏。
• ▸明确预测起点t与终点t+5→只让训练标签终点早于测试起点→冻结阈值→记录末训练和首测试时点。
• ▸本课作业：测试特征从168开始，标签跨度5，训练标签终点必须<168。训练特征最晚可为多少？

Practical Application

After studying, complete the companion practice lesson to move from understanding to doing.

• ▸Follow along in Jupyter or VS Code
• ▸Log how parameter changes affect backtest results
• ▸Submit a reproducible analysis notebook

FAQ

• ▸Q: How does this lesson fit the course? A: Part of "特征工程"—follow the curriculum order.
• ▸Q: What do I need? A: Demo account and notes; quant lessons need a Python environment.
• ▸Q: How do I know I've mastered it? A: Pass the quiz and explain three key points of 标签设计 on your own.