Adding more Learning-Based Repair tools to Cerberus

[Marti2203]

There is a list of possible tools one can work on adding to Cerberus in order to increase the breadth of the framework.

• KNOD
• CURE
• CoCoNut
• VRepair
• RewardRepair

[thanhlecongg]

Hi @Marti2203, did you find trained models for KNOD and CoCoNUT? I'm working on add more learning-based repair to Cerberus but I cannot find trained models for these techniques.

[Marti2203]

Hi @thanhlecongg , I unfortunately have not found CoCoNut's model yet, was searching for it last week. I have to check again for KNOD

[thanhlecongg]

Yeah, me too. For KNOD, I also cannot find the models. However, I found there are two other techniques with trained models:

• SelfAPR: [Code] [Model]
• AlphaRepair: [Code] (This technique uses CodeBERT directly so it is available in hugging face)

Btw, Do you plan to add any learning-based approaches in the near future? I also plan to do that so we can share to avoid wasting efforts.

[Marti2203]

I just checked for KNOD - yeah, I also cannot find the model. Thought that the docker contained it... I am unfortunately not going to add more learning-based tools soon as I am helping for APR-COMP :D Will message you if I am planning to add a tool

[thanhlecongg]

Ok. No worries. I also planned to do it. I will contribute after I complete.

[Marti2203]

Thanks a lot! :)

[Marti2203]

Will add you to the assigned people to allow for easier tracking

[thanhlecongg]

AlphaRepair has been added by Pull request #100. I will keep adding more learning-based repair tools.

[thanhlecongg]

SelfAPR has been added by Pull request #103.

[thanhlecongg]

@Marti2203 I'm working on RewardRepair. Unfortunately, they do not provide preprocessing code for their work. I have tried to reach the authors but have not received their response until now. I have implemented my own preprocessing code following their descriptions in the paper but the processed inputs are not identical to their provided data. Should I make a pull request for RewardRepair?

[Marti2203]

You can do a PR, but add an I to the name so there is some way of tracking that it is different

[thanhlecongg]

Ok, I got it.

[thanhlecongg]

RewardRepair has been added by Pull request #104.

[thanhlecongg]

RewardRepair, SelfAPR, and AlphaRepair are all the tools that I can find the trained model until now. So I will pause this task.