search

Haiyao-Nero / InvariantStock

18 stars 7 forks source link

How to reproduce the result in the paper? #1

Closed muhoushaonian closed 3 weeks ago

muhoushaonian commented 3 weeks ago

Your article is very engaging. How can I reproduce the results you achieved in the Chinese stock market? I found that the IC and RankIC performance are not good in the dataset I constructed using 128 factors for the CSI 300.

Haiyao-Nero commented 3 weeks ago

The dataset used in this study was collected using Tushare. To reproduce the results, you will need to gather the data independently, as it is not a public dataset. However, we have included some sample data in this repository. Details of the stock features are provided in Tables 1 and 5.

muhoushaonian commented 3 weeks ago

Thanks for your reply. By the way, I' curious about the difference between features dimension and number of factors in args setting. For example, you select 26 features for China Stock Market, how to set these two args?

Haiyao-Nero commented 3 weeks ago

Since the predictor is a typical FactorVAE structure, the factors are the output of the FactorVAE's encoder.

muhoushaonian commented 3 weeks ago

Thanks for your reply.