Problems when reproducing your results.

[jiangpw41]

I reproduced your experiment, but the results were not good.

Firstly, due to the fact that in the demonstration code you provided, only a single corresponding CSV file was selected for each test sample in its class dataset, such as the 'dataset/nba/data/verrtex_layer. csv' file in the code example (which does not actually exist in your open-source dataset). But in fact, you did not inform me how to accurately select this CSV file as the search object, and the evaluation was conducted in the entire class space. Therefore, I simply modified your code to make the search object the entire class space. Given this premise, we trained ChatGLM3-6B on your training set for 4 epochs as the ranker. Other factors remain unchanged, but in the final result, CA is only 32.70, far lower than the 91.57 you provided; In addition, since you did not provide a database file that can be directly executed, I originally planned to import the database using your CSV file. But the above CA results forced me to stop reproducing. If you could provide some explanations, I would be extremely grateful.