Why the models are trained multiple times?

[tbbatbb]

Hi.

I have 2 questions about the code:

1. Models are trained at the line 32 of the file 3.classifier/multi-label_classifier/find_best_para/find_best_parameters_for_models.py, but they are trained again at line 125 of the file 4.apply_classifier_to_test_projects/multi-label_classifiers/using_multi-label_classifiers.py. The two training process both iterates over all 6 models. Is this necessary?

2. It seems that the released code is only about generating SFS. I'm wondering about how to use the generated SFSs for source-to-binary function matching? The paper "https://arxiv.org/pdf/2210.15159" claims that O2NMatcher is integrated with CodeCMR/BinaryAI to perform binary2source function matching, is there any sample code demonstrating how to integrate the two tools?

[island255]

1. Models trained at the file 3.classifier/multi-label_classifier/find_best_para/find_best_parameters_for_models.py aims to find the best parameter for these models. Strictly, the dataset used to find the best parameter should not include the testing dataset in 4.apply_classifier_to_test_projects/multi-label_classifiers/using_multi-label_classifiers.py. Thus these two training processes should be different.
2. This work was conducted during my internship at Tencent, where I can access their internal model. Now I am not familiar with their UI, I do not know if the SFSs can be uploaded to detect similarities with binary functions.