KaggleHub doesn't unpack zip files on Windows

[MichalPuzanov]

Zip issue on Windows

I am using KaggleHub in Python. The issue is that Kagglehub is able to download a particular file from a zip achieve. This works well in Linux, but in Windows, it downloads the whole archive. I believe it is the same functionality as on Linux, but then it tries to unzip the archive and select the file and remove the others. But I'm not sure about it. The issue is unzip is not in Windows. Another possibility is that it doesn't want to unzip the file with a different extension. It expects zip to unzip it, but has for example csv.

The example can be this:

dataset: "datasnaek/youtube-new"
file: "GBvideos.csv"

videos_path = kh.dataset_download(config['kaggle']['videos']['dataset'], path=config['kaggle']['videos']['file'])

On Linux particular CSV file is downloaded to the cache, but on Windows, there is a whole zip archive in the cache.

I know a simple solution would be doesn't use Windows crap :-), but I have to. Is there any way around except of secure it in the code?

[VitriolOlja]

I ran into a related issue with kagglehub when downloading individual .npz files from this dataset.

Specifically, when I used kagglehub to download "k49-train-imgs.npz" and "k49-train-labels.npz", the first file (imgs) turned out to be a nested .npz — i.e., np.load() returned a bytes object instead of an array. I had to unpack it like this:

with np.load(filepath, allow_pickle=True) as data:
        key = data.files[0]
        value = data[key]  # value was of type bytes
        with np.load(io.BytesIO(value), allow_pickle=True) as inner_data:
                   inner_key = inner_data.files[0]
                   return inner_data[inner_key]

This did not happen when I downloaded the full ZIP manually and extracted the files myself. So something happened here when downloading that .npz file.

Also, as a workaround for the Windows unzip/cache issue, I temporarily changed the cache directory like so:

import os
import tempfile

original_cache = os.environ.get("KAGGLEHUB_CACHE")
with tempfile.TemporaryDirectory() as temp_dir:
    os.environ["KAGGLEHUB_CACHE"] = temp_dir
    # download and copy desired files
os.environ["KAGGLEHUB_CACHE"] = original_cache

This way I just got the files I wanted - and the rest are deleted. Safest is probably to also put it in a try/finally to put the environment variable back to what it was.

[MichalPuzanov]

I think it would be slightly more difficult with zip archive. My point is, that kagglehub works on Linux but not on Windows. So there are two possibilities, repair it on Windows or change documentation.