Dshiriaev

[dm-shr]

added evaluation tools in evaluation/rag_evaluate:

1. evaluate_rag_utils.py
2. generate_questions.ipynb to show how to generate Q-A pairs ('doc_question_dataset_1000.json', etc) to the provided index.
3. irrelevant_words.txt, relevant_words.txt - keywords for sampling more useful nodes when sampling nodes for Q-A pairs.
4. gpt_questions.txt, with/out prompt - GPT generated questions within data science: 70 - analytics, 50 - ML general, 50 - ranking, pricing, deployment, etc (ML topics), 20 - time series, 20 - MLOps, data engineering.
5. rag_evaluate.ipynb to show how to evaluate the retrieval/answering by generating 'rag_evaluate_json_512_postprocess_both_1.json' and 'rag_evaluate_json_1024.json' sample datasets.