from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Features, Value, ClassLabel, Image, Sequence
import csv
import datasets
import ast

class CAFOSatConfig(datasets.BuilderConfig):
    def __init__(self, split_column="cafosat_set1_training_train", **kwargs):
        super().__init__(**kwargs)
        self.split_column = split_column

class CAFOSat(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        CAFOSatConfig(name="set1_train", split_column="cafosat_set1_training_train", description="Set 1 training split"),
        CAFOSatConfig(name="set1_val", split_column="cafosat_set1_training_val", description="Set 1 validation split"),
        CAFOSatConfig(name="verified_train", split_column="cafosat_verified_training_train", description="Verified training split"),
        CAFOSatConfig(name="all_train", split_column="cafosat_all_training_train", description="Verified training split"),
    ]
    DEFAULT_CONFIG_NAME = "all_train"

    def _info(self):
        return DatasetInfo(
            description="CAFOSat: Remote sensing CAFO dataset with bounding boxes and infrastructure tags.",
            features=Features({
                "patch_file": Image(),
                "label": ClassLabel(
                    names=["Negative", "Swine", "Dairy", "Beef", "Poultry", "Horses", "Sheep/Goats"]
                ),
                "barn": Value("float32"),
                "manure_pond": Value("float32"),
                "grazing_area": Value("float32"),
                "others": Value("float32"),
                "geom_bbox": Sequence(Value("float32")),  # Keep as raw list
                "category": Value("string"),
                "state": Value("string"),
                "image_type": Value("string"),
                "CAFO_UNIQUE_ID": Value("string"),
                "verified_label": Value("string"),
                "patch_res": Value("string")
            }),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/oishee3003/CAFOSat/",
            license="cc-by-4.0"
        )

    def _split_generators(self, dl_manager):
        csv_path = dl_manager.download_and_extract("cafosat.csv")
        return [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"csv_path": csv_path, "split_flag": self.config.split_column})
        ]

    def _generate_examples(self, csv_path, split_flag):
        with open(csv_path, encoding="utf-8") as f:
            reader = csv.DictReader(f)
            for idx, row in enumerate(reader):
                if row.get(split_flag, "0") != "1":
                    continue

                # Parse bbox without scaling
                try:
                    bbox = ast.literal_eval(row.get("geom_bbox", "[5.0, 5.0, 700.0, 700.0]"))
                except:
                    bbox = [5.0, 5.0, 700.0, 700.0]

                yield idx, {
                    "patch_file": row["patch_file"],
                    "label": int(row["label"]),
                    "barn": float(row.get("barn", 0)),
                    "manure_pond": float(row.get("manure_pond", 0)),
                    "grazing_area": float(row.get("grazing_area", 0)),
                    "others": float(row.get("others", 0)),
                    "geom_bbox": bbox,  # ✅ unchanged
                    "category": row.get("category", ""),
                    "state": row.get("state", ""),
                    "image_type": row.get("image_type", ""),
                    "CAFO_UNIQUE_ID": row.get("CAFO_UNIQUE_ID", ""),
                    "verified_label": row.get("verified_label", ""),
                    "patch_res": row.get("patch_res", "")
                    "refine_x": row.get("refine_x", "")
                    "refine_y": row.get("refine_y", "")
                }