How to promote generalization ability of DEAP genetic programming?

[张尧中]

I want to apply genetic programming in stock data. The output function of genetic programming usually perform well on in-sample data but really bad on out-of-sample data. Can anybody give me some advice to promote generalization ability of DEAP genetic programming?

[Derek Tishler]

Sounds like a general overfitting problem for a trading agent and will depend a great deal on your trading approach and goals. It is easy to overfit to stock data with things like too few/too many trades or other issues like bad fill model due to sparse data or bad assumptions in the backtester yielding unrealistic or unusable results.

One trick from this paper is to rely on the population itself and probe, quickly, for non-overfit items. Some might argue that is a great way to overfit... Either way, there may be a handful of ideas to encourage you in this paper:

Evolving intraday foreign exchange trading strategies utilizing multiple instruments price series

https://arxiv.org/pdf/1411.2153.pdf

[张尧中]

This paper inspire me a lot. Thank you very much.

在 2019年8月27日星期二 UTC+8下午3:27:22，Derek Tishler写道：