{ "id": 297213, "title": "Why internet option needs to keep off during inference on kaggle?", "url": "/competitions/tensorflow-great-barrier-reef/discussion/297213", "author_name": "", "post_date": "2021-12-26T04:23:55.559855900Z", "votes": 4, "comment_count": 7, "views": 0, "content": "

Hi, can someone explain to me why on Kaggle during inference Kaggle asks to keep off the internet option?

", "messages": [ { "id": "1629374", "postDate": "12/26/2021 04:23:55", "content": "

Hi, can someone explain to me why on Kaggle during inference Kaggle asks to keep off the internet option?

", "rawMarkdown": "Hi, can someone explain to me why on Kaggle during inference Kaggle asks to keep off the internet option?", "votes": null }, { "id": "1629387", "postDate": "12/26/2021 04:42:52", "content": "

I think this is because to prevent sending private test data to outside of the internet.
\nThe purpose of kernel competition is the host consider to hide the test data, being afraid of over fitting to test data or other reason. So keeping test data private is necessary.

", "rawMarkdown": "I think this is because to prevent sending private test data to outside of the internet.\nThe purpose of kernel competition is the host consider to hide the test data, being afraid of over fitting to test data or other reason. So keeping test data private is necessary.", "votes": null }, { "id": "1629398", "postDate": "12/26/2021 05:07:31", "content": "

@tatamikenn thanks for clearing my doubts. But test data is hidden and it cant be accessible by us, so how does it matter if the internet is ON or OFF. Because hidden test data is only evaluated on submission only.

", "rawMarkdown": "tatamikenn thanks for clearing my doubts. But test data is hidden and it cant be accessible by us, so how does it matter if the internet is ON or OFF. Because hidden test data is only evaluated on submission only.", "votes": null }, { "id": "1629551", "postDate": "12/26/2021 09:43:39", "content": "

Well, imagine you write a kernel to get a test data and send it to google drive (or something else).
\nIf you submit the kernel, and evaluated on the evaluation environment, then you will get the whole test data.

", "rawMarkdown": "Well, imagine you write a kernel to get a test data and send it to google drive (or something else).\nIf you submit the kernel, and evaluated on the evaluation environment, then you will get the whole test data.", "votes": null }, { "id": "1629553", "postDate": "12/26/2021 09:47:03", "content": "
\n

Because hidden test data is only evaluated on submission only.

\n
\n

Actually, it's not true if the kernel can access to the internet.
\nYou can get the test data on the evaluation envieonment using the below api:

\n
import greatbarrierreef\nenv = greatbarrierreef.make_env()   # initialize the environment\niter_test = env.iter_test()    # an iterator which loops over the test set and sample submission\nfor (pixel_array, sample_prediction_df) in iter_test:\n    sample_prediction_df['annotations'] = '0.5 0 0 100 100'  # make your predictions here\n    env.predict(sample_prediction_df)   # register your predictions\n
", "rawMarkdown": "> Because hidden test data is only evaluated on submission only.\n\nActually, it's not true if the kernel can access to the internet.\nYou can get the test data on the evaluation envieonment using the below api:\n\n```\nimport greatbarrierreef\nenv = greatbarrierreef.make_env() # initialize the environment\niter_test = env.iter_test() # an iterator which loops over the test set and sample submission\nfor (pixel_array, sample_prediction_df) in iter_test:\n sample_prediction_df['annotations'] = '0.5 0 0 100 100' # make your predictions here\n env.predict(sample_prediction_df) # register your predictions\n```", "votes": null }, { "id": "1629571", "postDate": "12/26/2021 10:08:46", "content": "

Maybe you are not clear about how kernel is evaluated on the kernel competition. Below link may be helpful: https://towardsdatascience.com/submitting-model-predictions-to-kaggle-competition-ccb64b17132e

", "rawMarkdown": "Maybe you are not clear about how kernel is evaluated on the kernel competition. Below link may be helpful: https://towardsdatascience.com/submitting-model-predictions-to-kaggle-competition-ccb64b17132e", "votes": null }, { "id": "1634778", "postDate": "01/01/2022 06:53:38", "content": "

@tatamikenn thank you I went through the link u shared that cleared my doubts.
\nThank You

", "rawMarkdown": "tatamikenn thank you I went through the link u shared that cleared my doubts.\nThank You", "votes": null }, { "id": "1635057", "postDate": "01/01/2022 12:45:23", "content": "

Your welcome. 👍

", "rawMarkdown": "Your welcome. 👍", "votes": null } ], "comments": [ { "id": 1629387, "author_name": "tatamikenn", "author_url": "", "post_date": "12/26/2021 04:42:52", "content": "

I think this is because to prevent sending private test data to outside of the internet.
\nThe purpose of kernel competition is the host consider to hide the test data, being afraid of over fitting to test data or other reason. So keeping test data private is necessary.

", "votes": null, "replies": [ { "id": 1629398, "author_name": "anshulkhadse", "author_url": "", "post_date": "12/26/2021 05:07:31", "content": "

@tatamikenn thanks for clearing my doubts. But test data is hidden and it cant be accessible by us, so how does it matter if the internet is ON or OFF. Because hidden test data is only evaluated on submission only.

", "votes": null, "replies": [] }, { "id": 1629551, "author_name": "tatamikenn", "author_url": "", "post_date": "12/26/2021 09:43:39", "content": "

Well, imagine you write a kernel to get a test data and send it to google drive (or something else).
\nIf you submit the kernel, and evaluated on the evaluation environment, then you will get the whole test data.

", "votes": null, "replies": [] }, { "id": 1629553, "author_name": "tatamikenn", "author_url": "", "post_date": "12/26/2021 09:47:03", "content": "
\n

Because hidden test data is only evaluated on submission only.

\n
\n

Actually, it's not true if the kernel can access to the internet.
\nYou can get the test data on the evaluation envieonment using the below api:

\n
import greatbarrierreef\nenv = greatbarrierreef.make_env()   # initialize the environment\niter_test = env.iter_test()    # an iterator which loops over the test set and sample submission\nfor (pixel_array, sample_prediction_df) in iter_test:\n    sample_prediction_df['annotations'] = '0.5 0 0 100 100'  # make your predictions here\n    env.predict(sample_prediction_df)   # register your predictions\n
", "votes": null, "replies": [] }, { "id": 1629571, "author_name": "tatamikenn", "author_url": "", "post_date": "12/26/2021 10:08:46", "content": "

Maybe you are not clear about how kernel is evaluated on the kernel competition. Below link may be helpful: https://towardsdatascience.com/submitting-model-predictions-to-kaggle-competition-ccb64b17132e

", "votes": null, "replies": [] }, { "id": 1634778, "author_name": "anshulkhadse", "author_url": "", "post_date": "01/01/2022 06:53:38", "content": "

@tatamikenn thank you I went through the link u shared that cleared my doubts.
\nThank You

", "votes": null, "replies": [] }, { "id": 1635057, "author_name": "tatamikenn", "author_url": "", "post_date": "01/01/2022 12:45:23", "content": "

Your welcome. 👍

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1629374": "Hi, can someone explain to me why on Kaggle during inference Kaggle asks to keep off the internet option?", "1629387": "I think this is because to prevent sending private test data to outside of the internet.\nThe purpose of kernel competition is the host consider to hide the test data, being afraid of over fitting to test data or other reason. So keeping test data private is necessary.", "1629398": "tatamikenn thanks for clearing my doubts. But test data is hidden and it cant be accessible by us, so how does it matter if the internet is ON or OFF. Because hidden test data is only evaluated on submission only.", "1629551": "Well, imagine you write a kernel to get a test data and send it to google drive (or something else).\nIf you submit the kernel, and evaluated on the evaluation environment, then you will get the whole test data.", "1629553": "> Because hidden test data is only evaluated on submission only.\n\nActually, it's not true if the kernel can access to the internet.\nYou can get the test data on the evaluation envieonment using the below api:\n\n```\nimport greatbarrierreef\nenv = greatbarrierreef.make_env() # initialize the environment\niter_test = env.iter_test() # an iterator which loops over the test set and sample submission\nfor (pixel_array, sample_prediction_df) in iter_test:\n sample_prediction_df['annotations'] = '0.5 0 0 100 100' # make your predictions here\n env.predict(sample_prediction_df) # register your predictions\n```", "1629571": "Maybe you are not clear about how kernel is evaluated on the kernel competition. Below link may be helpful: https://towardsdatascience.com/submitting-model-predictions-to-kaggle-competition-ccb64b17132e", "1634778": "tatamikenn thank you I went through the link u shared that cleared my doubts.\nThank You", "1635057": "Your welcome. 👍" }, "source": "meta" }