search

SJTU-Quant / MASTER

This is the official code and supplementary materials for our AAAI-2024 paper: MASTER: Market-Guided Stock Transformer for Stock Price Forecasting. MASTER is a stock transformer for stock price forecasting, which models the momentary and cross-time stock correlation and guide feature selection with market information.
98 stars 20 forks source link

Inconsistent master dataset handling #8

Closed caozhiy closed 4 days ago

caozhiy commented 2 weeks ago

Thank you for your fancy job!

Recently when I wanted to use Qlib to get a dataset in Master format for the latest time, I found that the Label handling method integrated in Qlib was inconsistent with the method provided by the author in Issue. The CSRankNorm is used for Label in Qlib, but the author provides CSZScoreNorm to process Label.

May I ask which method is correct?

Best wishes!

Handler provided by author image

Qlib Handler image

LITONG99 commented 2 weeks ago

Thanks for pointing out the inconsistency to us.

To align with our experiments, please use CSZScoreNorm as we did, which is also specified in our paper. You can simply alter the .yaml configuration.

If there are other conflicts, please refer to the published paper and the content in this repository for clarification.

For discussion, if you use the CSRankNorm, the label is the normalized rank instead of the normalized return ratio. It is reasonable to use either of the two settings to train the MASTER model and the reported metrics will be ‘correct’ in the corresponding setting.