About the baseline of Java any-code completion

[ShangwenWang]

Dear authors,
Thanks for your work!
Your work is really great since there is no any-code completion tool before. I wonder would you release the baseline you designed to compare with AnyCodeGen on Java language (e.g., the retrained code2seq). I find building and retraining them would cause lots of time and energy. Thanks.

[ShangwenWang]

I mean can I query the trained code2seq like this
Really need your help. Thanks

[urialon]

Hi @ShangwenWang,
Thank you for your interest in our paper!

This baseline was a prototype adaptation of code2seq to SLM by taking the data of SLM and converting it to the code2seq format.

The head_paths from SLM are the paths that we used in code2seq.
The target sequence (that should be predicted) in code2seq is the linearized_tree in SLM or target_seq (depending on whether you wish to predict the target as a sequence of tokens or a linearized tree).

Best,
Uri

[ShangwenWang]

Dear @urialon
Thanks for your reply. I wonder will you provide a query link to the adaptation just like you have done for AnyCodeGen.
It is time consuming if retraining the model, not to mention any technical problem I might meet.