CanadianInvertebrates-ML.py

import csv
import datasets

class DNABarcodeDataset(datasets.GeneratorBasedBuilder):
    def _info(self):
        return datasets.DatasetInfo(
            description="DNA barcode dataset with hierarchical taxonomy labels and multiple splits.",
            features=datasets.Features({
                "processid": datasets.Value("string"),
                "sampleid": datasets.Value("string"),
                "dna_bin": datasets.Value("string"),
                "phylum": datasets.Value("string"),
                "class": datasets.Value("string"),
                "order": datasets.Value("string"),
                "family": datasets.Value("string"),
                "genus": datasets.Value("string"),
                "species": datasets.Value("string"),        # label
                "dna_barcode": datasets.Value("string"),    # input data
                "split": datasets.ClassLabel(names=["train", "val", "test", "test_unseen", "pretrain"]),
            }),
            supervised_keys=("dna_barcode", "species"),  # For model training
        )

    def _split_generators(self, dl_manager):
        data_path = dl_manager.download("https://huggingface.co/datasets/bioscan-ml/CanadianInvertebrates-ML/resolve/main/CanInv_metadata.csv")  # Use a URL or relative path
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_path, "split": "train"}),
            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": data_path, "split": "val"}),
            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": data_path, "split": "test"}),
            datasets.SplitGenerator(name="test_unseen", gen_kwargs={"filepath": data_path, "split": "test_unseen"}),
            datasets.SplitGenerator(name="pretrain", gen_kwargs={"filepath": data_path, "split": "pretrain"}),
        ]

    def _generate_examples(self, filepath, split):
        with open(filepath, encoding="utf-8") as f:
            reader = csv.DictReader(f)
            idx = 0
            for row in reader:
                if row["split"] == split:
                    yield idx, {
                        "processid": row["processid"],
                        "sampleid": row["sampleid"],
                        "dna_bin": row["dna_bin"],
                        "phylum": row["phylum"],
                        "class": row["class"],
                        "order": row["order"],
                        "family": row["family"],
                        "genus": row["genus"],
                        "species": row["species"],
                        "dna_barcode": row["dna_barcode"],
                        "split": row["split"],
                    }
                    idx += 1