`competition_download` does not download to standard path

[Borda]

I am running a Google Colab session generated by Kaggle -> open in Colab
The notebook has a competition dataset and additional users' datasets, so the user datasets are downloaded as expected to the folder kaggle/inputs, but the competition dataset stays in cache only, see this sample code:

cafa_6_protein_function_prediction_path = kagglehub.competition_download('cafa-6-protein-function-prediction')
sergeifironov_t5embeds_path = kagglehub.dataset_download('sergeifironov/t5embeds')
viktorfairuschin_cafa_5_ems_2_embeddings_numpy_path = kagglehub.dataset_download('viktorfairuschin/cafa-5-ems-2-embeddings-numpy')
henriupton_protbert_embeddings_for_cafa5_path = kagglehub.dataset_download('henriupton/protbert-embeddings-for-cafa5')
siddhvr_train_targets_top500_path = kagglehub.dataset_download('siddhvr/train-targets-top500')
siddhvr_blast_quick_sprof_zero_pred_path = kagglehub.dataset_download('siddhvr/blast-quick-sprof-zero-pred')

print('Data source import complete.')
print(f'cafa_6_protein_function_prediction_path: {cafa_6_protein_function_prediction_path}')
print(f'sergeifironov_t5embeds_path: {sergeifironov_t5embeds_path}')
print(f'viktorfairuschin_cafa_5_ems_2_embeddings_numpy_path: {viktorfairuschin_cafa_5_ems_2_embeddings_numpy_path}')
print(f'henriupton_protbert_embeddings_for_cafa5_path: {henriupton_protbert_embeddings_for_cafa5_path}')
print(f'siddhvr_train_targets_top500_path: {siddhvr_train_targets_top500_path}')

with output:

Data source import complete.
cafa_6_protein_function_prediction_path: /root/.cache/kagglehub/competitions/cafa-6-protein-function-prediction
sergeifironov_t5embeds_path: /kaggle/input/t5embeds
viktorfairuschin_cafa_5_ems_2_embeddings_numpy_path: /kaggle/input/cafa-5-ems-2-embeddings-numpy
henriupton_protbert_embeddings_for_cafa5_path: /kaggle/input/protbert-embeddings-for-cafa5
siddhvr_train_targets_top500_path: /kaggle/input/train-targets-top500

happy to address this issue

cc: @rosbo @mohami2000 @goeffthomas

[rosbo]

This is working as intended.

Colab has a shared dataset cache (similar to Kaggle) and that shared cache is mounted at /kaggle/input.

However, Colab cache only supports Kaggle datasets and not Kaggle competition datasets which is why the competition dataset is downloaded from Kaggle and stored in a local cache folder (/root/.cache/kagglehub/competitions). I can check with the Colab team whether they would consider caching also Kaggle competition datasets.

Note: Kaggle competition datasets are slightly different than a regular public dataset because you need to join the competition & accept the terms and conditions before you can access the data. This means you would need to make an authenticated call using kagglehub to access a competition dataset so we can ensure you have indeed joined the competition.

If you use the path returned by kagglehub in your code, do you run into issues?

Thank you