libsvm.py

from benchopt import BaseDataset, safe_import_context

with safe_import_context() as import_ctx:
    from libsvmdata import fetch_libsvm

    preprocess_data = import_ctx.import_from("utils", "preprocess_data")


class Dataset(BaseDataset):

    name = "libsvm"

    parameters = {
        "dataset": ["finance", "finance-tf-idf", "YearPredictionMSD"],
        "standardize" : [True, False]
    }

    install_cmd = "conda"
    requirements = ["pip:libsvmdata"]

    def __init__(self, dataset="bodyfat", standardize=True):
        self.dataset = dataset
        self.X, self.y = None, None
        self.standardize = standardize

    def get_data(self):
        X, y = fetch_libsvm(self.dataset)
        if self.standardize:
            X, y = preprocess_data(X, y)

        return dict(X=X, y=y)