Decoupling Qlib and AlphaForge #20
Description
Congratulations on your paper being accepted by the top committee! The idea in your paper is very innovative.
The input for AlphaForge's strong coupling with Qlib is complex, making it difficult to replace Qlib and limiting the scalability of the program.
In my opinion, Qlib only provides daily stock DataFrame data in the program. Therefore, I suggest saving the original DataFrame to local .pkl files, which users can then read using pd.read_pickle(). Users can easily change the data source, as long as they ensure the new source follows the .pkl file arrangement and naming rules specified by the program.
However, the current program loads the original data from Qlib, places it into the StockData class, and then saves this StockData class to a .pkl file in stock_data.py. The subsequent training steps rely on these .pkl files, which cannot be easily read by pd.read_pickle(). Therefore, when users want to replace Qlib with a different data source, the renovation cost is very high.