dataset = tf.data.experimental.make_csv_dataset(
    "training.csv",
    batch_size=2,
    label_name="label",
    num_epochs=1,
    shuffle=False,
)