When it comes to feature engineering, possibilities are seemingly limitless, and there is certainly room for some experimentation and creativity. The good news is there are already quite a few packages out there that do the job for us, like meteostat for historical weather data or yfinance for stock market data. The bad one is, there is generally no clear answer to which additional features may improve the model performance without actually trying them out.
说到特征工程,可能的组合似乎是无限的,当然还有一些实验和创造力的空间。好消息是已经有相当多的软件包为我们做了这项工作,比如历史天气数据的meteostat或股票市场数据的yfinance。糟糕的是,通常没有明确的答案来说明哪些附加特征可以在不实际试用的情况下提高模型性能。
去掉“16个交易日内的涨停情况(riseStopInfo)”、“
利用captum.attr的IntegratedGradients工具进行特征选择实验
最新推荐文章于 2024-06-03 09:38:03 发布
本文探讨了在特征工程中如何利用captum.attr的IntegratedGradients工具进行特征选择。通过实验发现,移除某些特征如'16个交易日内的涨停情况'和'跌停情况'会导致模型性能下降,而特征'day_K'的重要性较低,可能存在不准确性。通过模型压缩,可以处理更多股票的数据,从而提高计算效率。
摘要由CSDN通过智能技术生成