Open bell99431 opened 1 year ago
I'm using qlib library for model training. The data_handler.py extends the Alpha158
class and line14 set the predicting range to 9 days (10 - 1 = 9, although I was originally intended to set to 10 days).
If you want to predict N days, change 10 to N + 1 and re-train the model.
If you want to predict one day, just use Alpha158
instead of Alpha158TwoWeeks
.
Predicting for any days could be supported later, either as a config or generally supported.
I haven't implemented the backtest part of this model. You could refer to the backtest of one-day-models in qlib: https://github.com/microsoft/qlib/tree/main/examples/benchmarks#alpha158-dataset We are using LightGBM model.
Thx, I'll try it.
Cool! If you have any suggestions or improving for the code, please let me know. Thank you!
Thanks for share! How to config the predict as N days? And, What's the predict accurate data with backtest?