urialon
commented
4 years ago Hi @shreyasingh , thank you for your interest in code2vec!
- As I answered in another thread - you can try the raw dataset here: https://miltos.allamanis.com/publications/2018learning/
- Alternatively, you can just
git cloneas much data as you wish from public repositories. - We did not release the trained C# model.
- In general, see also https://code2seq.org/ which is a more powerful model than code2vec.
- @ASvyatkovskiy have used code2vec/seq for C# in the past, he might have trained models or data for you?
Best, Uri
Hi, I'm working on a Neural Code Search prototype and am currently in the process of doing literature survey on code embeddings. I came across your code2vec paper and I must say that it is a very well-written paper and I enjoyed reading it!
I was going through the github code for code2vec (https://github.com/tech-srl/code2vec) and wanted to train the model for C# files. Could you let me know if:
There is an existing dataset for C# source code files - the same way as you've published for Java on Github. The dataset could be unprocessed, and I could process it with preprocess_csharp.sh. I also saw that you've published the token and method name embeddings which are available to download. Is there a similar token and method name embedding file available for C# language tokens that I could directly load and use? That way I will not need to train the model. Your advice and guidance on the above two items would be highly appreciated.