How can I download only the train and test split for full numbers using load_dataset()?

[Nakkhatra]

How can I download only the train and test split for full numbers using load_dataset()?

I do not need the extra split and it will take 40 mins just to download in Colab. I have very short time in hand. Please help.

[mariosasko]

Hi! Can you please specify the full name of the dataset? IIRC full_numbers is one of the configs of the svhn dataset, and its generation is slow due to data being stored in binary Matlab files. Even if you specify a specific split, datasets downloads all of them, but we plan to fix that soon and only download the requested split.

If you are in a hurry, download the svhn script here, remove this code, and run:

from datasets import load_dataset
dset = load_dataset("path/to/your/local/script.py", "full_numbers")

And to make loading easier in Colab, you can create a dataset repo on the Hub and upload the script there. Or push the script to Google Drive and mount the drive in Colab.