RT: Been playing a bit on @numerai
A few things that seem to be working:
- extremely simple models
- distrusting validation (trust, but verify)
- optimizing for stability rather than high correlation (models that don’t flip from pos to neg corr frequently)
转发：在@numerai上玩了一点
以下几点似乎有效：
-非常简单的模型
-不信任验证（信任，但验证）
-优化稳定性而非高相关性（不经常从pos转换为neg的模型）