biovdb_1000.py

import pandas as pd
import datasets

CITATION = """
"""
#TODO: There must be other way to refer to file in the repo
CSV_URL = "https://huggingface.co/datasets/mwinn99/biovdb_1000/resolve/main/biovdb_1000.csv"

class Biovdb(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")
    
    def _info(self):
        return datasets.DatasetInfo(
            description="Test set of samples from GEO",
            citation=CITATION,
            homepage="",
            features=datasets.Features({
                'metadata': {col: datasets.Value('string') for col in pd.read_csv(CSV_URL, nrows=1).columns[:40]},
                'data': datasets.Sequence(datasets.Value('float32'))
            }),
        )

    def _split_generators(self, dl_manager):
        data_fname = dl_manager.download(CSV_URL)
        data = pd.read_csv(data_fname)
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data": data})
        ]
    
    def _generate_examples(self, data):
        data = data.infer_objects()
        metadata_columns = data.columns[:40]
        expression_columns = data.columns[40:]

        for idx, row in data.iterrows():
            yield idx, {
                "data": row[expression_columns].astype(float).values.tolist(),
                "metadata": row[metadata_columns].astype(str).to_dict()
            }

if __name__ == '__main__':
    ds = datasets.load_dataset("./biovdb_1000.py", split='train')