import ragas
from ragas.metrics import (
    faithfulness,
    answer_relevancy,
    context_precision,
    context_recall,
    context_relevancy,
    answer_correctness
)

from datasets import Dataset 
from ragas import evaluate
from langchain_openai import ChatOpenAI, OpenAIEmbeddings

ragas_metrics = [ faithfulness,
    answer_relevancy,
    context_precision,
    context_recall,
    context_relevancy,
    answer_correctness]

def ragas_evaluate(df_res, metrics, query_col = "Question", truth_ans_col = "Answer" ,
                       llm = ChatOpenAI(model="gpt-3.5-turbo"),
                       embed_model = OpenAIEmbeddings(model="text-embedding-3-small"),
                       raise_exceptions = True,
                       timeout=10, max_retries=3, max_wait=10,
                       is_async = True,
                        callbacks = None
                   ):

        data_samples = {
            'question': df_res[query_col].to_list(),
            'answer': df_res.rag_answer.to_list(),
            'contexts' : df_res.context.to_list(),
            'ground_truth': df_res[truth_ans_col].to_list()
        }

        run_config = ragas.RunConfig(timeout=timeout, max_retries=max_retries, max_wait=max_wait)

        dataset = Dataset.from_dict(data_samples)
   
        for idx, metric in enumerate(metrics):
            print(f">> {idx+1} : {metric.name}")
            score = evaluate(dataset, metrics= [metric], llm = llm,
                        embeddings= embed_model, raise_exceptions=raise_exceptions, run_config=run_config,
                             is_async=is_async, callbacks=callbacks)
            time.sleep(1)
            if idx == 0:
                df_score = score.to_pandas()
            else:
                df_score = df_score.merge(score.to_pandas()[['question',metric.name]], on = 'question', how = "left")

        return df_score


# for now showing with only faithfullness and answer_relevancy
## without any data point which may cause exception
df_eval = ragas_evaluate(df_response.iloc[list(range(20))], ragas_metrics[0:2], raise_exceptions=False)
df_eval

## with a data point which cause exception - openai.BadRequestError: Error code: 400 [context_length_exceeded]
df_eval_ex = ragas_evaluate(df_res.iloc[list(range(20))+ [68]], ragas_metrics[0:2], raise_exceptions=False)
df_eval_ex