import re
import string
from lighteval.tasks.lighteval_task import LightevalTaskConfig
from lighteval.metrics import Metrics, MetricCategory
from lighteval.metrics.utils import CorpusLevelMetric, MetricUseCase
from aenum import extend_enum
import numpy as np
from lighteval.tasks.requests import Doc
from Levenshtein import distance
import collections
from lighteval.utils import as_list

def sentiment_eval_fn(golds: list[str], predictions: list[str], formatted_doc: Doc = None):
    if len(predictions)  > 1:
        raise ValueError("Predictions should have one item")
    # do some santizations, since some models produce more info
    pred = re.sub('<[^>]+>', '', predictions[0]) # remove xml tags
    pred = re.sub(r'^100%', '', pred) # remove 100% at beginning, some gemma weirdness
    pred = pred.strip()

    return 1 if pred == golds[0] else 0
    
sentiment_acc_metric = CorpusLevelMetric(
    metric="sentiment_acc",
    higher_is_better=True,
    category=MetricCategory.GENERATIVE,
    use_case=MetricUseCase.ACCURACY,
    corpus_level_fn=np.mean,
    sample_level_fn=sentiment_eval_fn
)
extend_enum(Metrics, 'sentiment_acc_metric', sentiment_acc_metric)

def sentiment_prompt_fn(line, task_name: str = None):
    """Defines how to go from a dataset line to a doc object.
    Follow examples in src/lighteval/tasks/tasks_prompt_formatting.py, or get more info
    about what this function should do in the README.
    """
    return Doc(
        task_name=task_name,
        query=line["prompt"],
        choices=line["response"],
        gold_index=0,
        instruction="",
    )

# This is how you create a simple tasks (like hellaswag) which has one single subset
# attached to it, and one evaluation possible.
sentiment_task = LightevalTaskConfig(
    name="sentiment-acc",
    prompt_function="sentiment_prompt_fn",  # must be defined in the file or imported from src/lighteval/tasks/tasks_prompt_formatting.py
    suite=["custom"],
    hf_repo="dicta-hebrew-llm-leaderboard/tests",
    hf_subset="default",
    hf_avail_splits=["sentiment"],
    evaluation_splits=["sentiment"],
    metric=['sentiment_acc_metric'],
    stop_sequence=['\n'],
    generation_size=32
)
sentiment_task.stop_sequence = as_list(sentiment_task.stop_sequence)