{
  "id": 100471,
  "title": "Public versus Private data set",
  "url": "/competitions/aptos2019-blindness-detection/discussion/100471",
  "author_name": "",
  "post_date": "2019-07-18T18:45:10.849486300Z",
  "votes": 1,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hello \nIs the kernel that we are now uploading and running for the public score, is running on the entire data set (and we only see the results of the reduced public set)?  In other words, the time it takes to run reflects the time for the entire data set? </p>",
  "messages": [
    {
      "id": "579338",
      "postDate": "07/18/2019 18:45:10",
      "content": "<p>Hello \nIs the kernel that we are now uploading and running for the public score, is running on the entire data set (and we only see the results of the reduced public set)?  In other words, the time it takes to run reflects the time for the entire data set? </p>",
      "rawMarkdown": "Hello \nIs the kernel that we are now uploading and running for the public score, is running on the entire data set (and we only see the results of the reduced public set)?  In other words, the time it takes to run reflects the time for the entire data set?",
      "votes": null
    },
    {
      "id": "579365",
      "postDate": "07/18/2019 19:12:56",
      "content": "<p>My kernel takes about ~ 2mins on public test set, after submitting it takes ~30 mins, I think it runs on entire private test data. At the end of the competition we have to select our best submissions, I don't think they will run that version of kernel again to obtain the final score, as the then added datasets might get deleted/altered in the meantime. That's one explanation.</p>",
      "rawMarkdown": "My kernel takes about ~ 2mins on public test set, after submitting it takes ~30 mins, I think it runs on entire private test data. At the end of the competition we have to select our best submissions, I don't think they will run that version of kernel again to obtain the final score, as the then added datasets might get deleted/altered in the meantime. That's one explanation.",
      "votes": null
    },
    {
      "id": "579595",
      "postDate": "07/19/2019 02:25:39",
      "content": "<p>The nine-hour GPU runtime limitation is for public test set(~ 2mins) or submitting(~30 mins)?</p>",
      "rawMarkdown": "The nine-hour GPU runtime limitation is for public test set(~ 2mins) or submitting(~30 mins)?",
      "votes": null
    },
    {
      "id": "579644",
      "postDate": "07/19/2019 03:53:42",
      "content": "<p>Nope. Kernel submission has nothing to do with that. That 9 hour is how long you can use Kaggle's kernels for training and experimentation.</p>",
      "rawMarkdown": "Nope. Kernel submission has nothing to do with that. That 9 hour is how long you can use Kaggle's kernels for training and experimentation.",
      "votes": null
    },
    {
      "id": "579846",
      "postDate": "07/19/2019 09:56:37",
      "content": "<p><a href=\"/omershect\">@omershect</a> what you said was already correct. Commit version runs on public only. Submit version runs on entire public+private.</p>",
      "rawMarkdown": "omershect what you said was already correct. Commit version runs on public only. Submit version runs on entire public+private.",
      "votes": null
    },
    {
      "id": "580157",
      "postDate": "07/19/2019 18:45:35",
      "content": "<blockquote>\n  <p>When Kaggle runs your Kernel privately, it substitutes the private test set and sample submission in place of the public ones. You can plan on the private test set consisting of 20GB of data across 13,000 images (approximately).</p>\n</blockquote>\n\n<p>1928 Public vs ~13k Privat. As you can calculate time for all set images ~7-8 time than public test images</p>",
      "rawMarkdown": "&gt; When Kaggle runs your Kernel privately, it substitutes the private test set and sample submission in place of the public ones. You can plan on the private test set consisting of 20GB of data across 13,000 images (approximately).\n\n1928 Public vs ~13k Privat. As you can calculate time for all set images ~7-8 time than public test images",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 579365,
      "author_name": "rishabhiitbhu",
      "author_url": "",
      "post_date": "07/18/2019 19:12:56",
      "content": "<p>My kernel takes about ~ 2mins on public test set, after submitting it takes ~30 mins, I think it runs on entire private test data. At the end of the competition we have to select our best submissions, I don't think they will run that version of kernel again to obtain the final score, as the then added datasets might get deleted/altered in the meantime. That's one explanation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 579595,
          "author_name": "dldmw579",
          "author_url": "",
          "post_date": "07/19/2019 02:25:39",
          "content": "<p>The nine-hour GPU runtime limitation is for public test set(~ 2mins) or submitting(~30 mins)?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 579644,
          "author_name": "rishabhiitbhu",
          "author_url": "",
          "post_date": "07/19/2019 03:53:42",
          "content": "<p>Nope. Kernel submission has nothing to do with that. That 9 hour is how long you can use Kaggle's kernels for training and experimentation.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 579846,
      "author_name": "ratthachat",
      "author_url": "",
      "post_date": "07/19/2019 09:56:37",
      "content": "<p><a href=\"/omershect\">@omershect</a> what you said was already correct. Commit version runs on public only. Submit version runs on entire public+private.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 580157,
      "author_name": "leighplt",
      "author_url": "",
      "post_date": "07/19/2019 18:45:35",
      "content": "<blockquote>\n  <p>When Kaggle runs your Kernel privately, it substitutes the private test set and sample submission in place of the public ones. You can plan on the private test set consisting of 20GB of data across 13,000 images (approximately).</p>\n</blockquote>\n\n<p>1928 Public vs ~13k Privat. As you can calculate time for all set images ~7-8 time than public test images</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "579338": "Hello \nIs the kernel that we are now uploading and running for the public score, is running on the entire data set (and we only see the results of the reduced public set)?  In other words, the time it takes to run reflects the time for the entire data set?",
    "579365": "My kernel takes about ~ 2mins on public test set, after submitting it takes ~30 mins, I think it runs on entire private test data. At the end of the competition we have to select our best submissions, I don't think they will run that version of kernel again to obtain the final score, as the then added datasets might get deleted/altered in the meantime. That's one explanation.",
    "579595": "The nine-hour GPU runtime limitation is for public test set(~ 2mins) or submitting(~30 mins)?",
    "579644": "Nope. Kernel submission has nothing to do with that. That 9 hour is how long you can use Kaggle's kernels for training and experimentation.",
    "579846": "omershect what you said was already correct. Commit version runs on public only. Submit version runs on entire public+private.",
    "580157": "&gt; When Kaggle runs your Kernel privately, it substitutes the private test set and sample submission in place of the public ones. You can plan on the private test set consisting of 20GB of data across 13,000 images (approximately).\n\n1928 Public vs ~13k Privat. As you can calculate time for all set images ~7-8 time than public test images"
  },
  "source": "meta"
}