- retest each version with new data.
- change the mapping of data -> because RAG only get value but key
- or fix it inside the RAG function.
THis is the final project form AI 2025, NTU CSIE:
for project detail please look at here
following is file description:
- each version of main code (v1 / v2 / v2-1 / v3&4 / v5 ) : each version of final project's test result, see detail here.
- threshold_test.py : to change probability into binary, with given threshold variable.
rule_build.py : for given .json files, turn them into three json file: law_text, legal_phrases and violation_phrase_map- now use the builder in data_builder/
- filter.py : if have multiple keys for output, this file can help change into kaggle competion's format.
- data/
- legal/illegal_phrases.json : only phrases.
- legal/illegal_phrases_map.json : map the reason with phrases.
- pip install -r requirement.txt
- ./data_builder/run.sh
- python3 v1/2/3&4.py
- python3 threshold_test.py
- in each version of code's main(), df means the query now use, so change to whether customized query you want.
- 5/25 start build the project
- 5/26 attempt v1 ~ v4
- 5/27 attempt v5 & v2-1
- 5/28 build phrases
- 5/29 build phrases