Questions about JavaExtractor's hash function about code2vec HOT 4 CLOSED

Hi,
Thank you for your interest in Code2vec.
I'm just letting you know that the encoder of code2seq is much better than code2vec's and does not use hashing. The implementation and extending to other languages is very similar.

Regarding your questions:

1. The hash function was simply Java's String#hashCode().
2. No, there is no need to unhash.
3. I am guessing that the arrows contribute only a few additional points. You can definitely drop them as a first step. In our PLDI'18 paper - without arrows the results were very similar (for other tasks and another language though)

Best,
Uri

from code2vec.

Yes! It is only intended to shorten the long path strings into shorter strings.

from code2vec.

Thank you for your prompt reply!

from code2vec.

So ideally I should be okay with using any hash function then?

from code2vec.