import pandas as pd
import os
from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Value, Features, Image

class TunisianProverbsDataset(GeneratorBasedBuilder):
    def _info(self):
        return DatasetInfo(
            description="Tunisian Proverbs with Multiple Image Associations",
            features=Features({
                "tunisan_proverb": Value("string"),
                "proverb_arabic_explaination": Value("string"),
                "context": Value("string"),
                "caption": Value("string"),
                "caption_formal": Value("string"),
                "dynamic": Value("string"),
                "prompt": Value("string"),
                "image_path_1": Image(),
                "image_path_2": Image(),
                "image_path_3": Image(),
                "image_path_4": Image(),
                "clip_scores": Value("float32")
            }),
        )

    def _split_generators(self, dl_manager):
        data_path = dl_manager.download_and_extract("dataset.csv")
        image_dir = dl_manager.download_and_extract("images/")
        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={
                    "csv_path": data_path,
                    "image_dir": image_dir
                }
            )
        ]

    def _generate_examples(self, csv_path, image_dir):
        df = pd.read_csv(csv_path)

        def safe_image_path(path):
            if pd.isna(path) or not isinstance(path, str) or not path.strip():
                return None
            full_path = os.path.join(image_dir, path.strip())
            return full_path if os.path.isfile(full_path) else None

        for idx, row in df.iterrows():
            yield idx, {
                "tunisan_proverb": row.get("tunisan_proverb", ""),
                "proverb_arabic_explaination": row.get("proverb_arabic_explaination", ""),
                "context": row.get("context", ""),
                "caption": row.get("caption", ""),
                "caption_formal": row.get("caption_formal", ""),
                "dynamic": row.get("dynamic", ""),
                "prompt": row.get("prompt", ""),
                "image_path_1": safe_image_path(row.get("image_path_1")),
                "image_path_2": safe_image_path(row.get("image_path_2")),
                "image_path_3": safe_image_path(row.get("image_path_3")),
                "image_path_4": safe_image_path(row.get("image_path_4")),
                "clip_scores": float(row["clip_scores"]) if pd.notna(row.get("clip_scores")) else 0.0,
            }