ztw33/precos

Precos: Project-specific Retrieval for Better Code Summarization

dependencies

Please refer to knn-transformers. Additionally, you should install pylucene for subset retrieval.

subset retrieval

cd to subset_retrieval, set LANG, TEST_PROJECT, KEY, and directory in syntax_subset_retrieval.sh and semantics_subset_retrieval.sh, then run them to perform sentence-level retrieval in history-based corpus construction. The output is a file containing the ids in CSN train set. Merge the four files to get the union of all-type retrieved ids. And then construct the ext-subset from CSN and ids.

datastore creation

First combine the retrieved ext-subset with project train set. Then set the arguments in datastore_creation.sh and run bash datastore_creation.sh to create datastore.

generation

locality index file

First separately perform datastore creation for ext-subset and project train set. You can get dstore_size.txt in dstore_dir. Then run python get_locality_index.py to get the locality index file.

generate summaries

Set the arguments in generation.sh and run bash generation.sh to generate summaries.

acknowledgement

The implementation is based on knn-transformers.