{
  "id": 116739,
  "title": "What will be done about data leaks?",
  "url": "/competitions/ashrae-energy-prediction/discussion/116739",
  "author_name": "",
  "post_date": "2019-11-11T06:35:12.011912400Z",
  "votes": 24,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Kaggle has already mentioned that they are aware of data leaks but they haven't mentioned how they will handle them. </p>\n\n<p>Will leaked segments of the data be removed from the private LB? I think it's important to know so we can work out what data to use in our models. </p>",
  "messages": [
    {
      "id": "670185",
      "postDate": "11/11/2019 06:35:12",
      "content": "<p>Kaggle has already mentioned that they are aware of data leaks but they haven't mentioned how they will handle them. </p>\n\n<p>Will leaked segments of the data be removed from the private LB? I think it's important to know so we can work out what data to use in our models. </p>",
      "rawMarkdown": "Kaggle has already mentioned that they are aware of data leaks but they haven't mentioned how they will handle them. \n\nWill leaked segments of the data be removed from the private LB? I think it's important to know so we can work out what data to use in our models.",
      "votes": null
    },
    {
      "id": "670354",
      "postDate": "11/11/2019 11:13:29",
      "content": "<p>This explains why some people jumped suddenly to the top of LB.\nIt has started and this is not gonna stop. LB will become useless.</p>",
      "rawMarkdown": "This explains why some people jumped suddenly to the top of LB.\nIt has started and this is not gonna stop. LB will become useless.",
      "votes": null
    },
    {
      "id": "670552",
      "postDate": "11/11/2019 15:12:00",
      "content": "<p>In general everyone should ask himself what are his goals at Kaggle?  For me it's better to spend time to becoming smarter in EDA, ML but looking for the leaks and exploit them. </p>",
      "rawMarkdown": "In general everyone should ask himself what are his goals at Kaggle?  For me it's better to spend time to becoming smarter in EDA, ML but looking for the leaks and exploit them.",
      "votes": null
    },
    {
      "id": "670884",
      "postDate": "11/12/2019 01:47:45",
      "content": "<p>In my opinion, the competition data should be limited within Kaggle only with no reference to external dataset so that it is fair for every participant.</p>",
      "rawMarkdown": "In my opinion, the competition data should be limited within Kaggle only with no reference to external dataset so that it is fair for every participant.",
      "votes": null
    },
    {
      "id": "671290",
      "postDate": "11/12/2019 13:42:32",
      "content": "<p>In the Humpback whale competition, the LB was altered to remove a small leak:\n<a href=\"https://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664\">https://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664</a></p>\n\n<p>Maybe the same can be done here? </p>\n\n<p>It's also interesting to see how different the first half of the train data is compared to the test data for the UCF site. This has a pretty big impact on what models for this site could look like.</p>",
      "rawMarkdown": "In the Humpback whale competition, the LB was altered to remove a small leak:\nhttps://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664\n\nMaybe the same can be done here? \n\nIt's also interesting to see how different the first half of the train data is compared to the test data for the UCF site. This has a pretty big impact on what models for this site could look like.",
      "votes": null
    },
    {
      "id": "672233",
      "postDate": "11/13/2019 17:03:01",
      "content": "<p>Is it not a good sign that using the leaked data improves the LB score - this suggests that at least some or most of the leaked data is not part of the private score.</p>",
      "rawMarkdown": "Is it not a good sign that using the leaked data improves the LB score - this suggests that at least some or most of the leaked data is not part of the private score.",
      "votes": null
    },
    {
      "id": "680376",
      "postDate": "11/24/2019 14:46:26",
      "content": "<p>I agree.The host should have restricted use of external data.This has now become a webscraping competition.✋ </p>",
      "rawMarkdown": "I agree.The host should have restricted use of external data.This has now become a webscraping competition.✋",
      "votes": null
    },
    {
      "id": "680380",
      "postDate": "11/24/2019 14:49:07",
      "content": "<p>How can you detect the person uses it or not? Someone can easily slip leaks to their submissions like a blend and no one will notice.</p>",
      "rawMarkdown": "How can you detect the person uses it or not? Someone can easily slip leaks to their submissions like a blend and no one will notice.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 670354,
      "author_name": "jielu0728",
      "author_url": "",
      "post_date": "11/11/2019 11:13:29",
      "content": "<p>This explains why some people jumped suddenly to the top of LB.\nIt has started and this is not gonna stop. LB will become useless.</p>",
      "votes": null,
      "replies": [
        {
          "id": 672233,
          "author_name": "pcjimmmy",
          "author_url": "",
          "post_date": "11/13/2019 17:03:01",
          "content": "<p>Is it not a good sign that using the leaked data improves the LB score - this suggests that at least some or most of the leaked data is not part of the private score.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 670552,
      "author_name": "andrejvetrov",
      "author_url": "",
      "post_date": "11/11/2019 15:12:00",
      "content": "<p>In general everyone should ask himself what are his goals at Kaggle?  For me it's better to spend time to becoming smarter in EDA, ML but looking for the leaks and exploit them. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 670884,
      "author_name": "klloke",
      "author_url": "",
      "post_date": "11/12/2019 01:47:45",
      "content": "<p>In my opinion, the competition data should be limited within Kaggle only with no reference to external dataset so that it is fair for every participant.</p>",
      "votes": null,
      "replies": [
        {
          "id": 680376,
          "author_name": "shahules",
          "author_url": "",
          "post_date": "11/24/2019 14:46:26",
          "content": "<p>I agree.The host should have restricted use of external data.This has now become a webscraping competition.✋ </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 680380,
          "author_name": "gunesevitan",
          "author_url": "",
          "post_date": "11/24/2019 14:49:07",
          "content": "<p>How can you detect the person uses it or not? Someone can easily slip leaks to their submissions like a blend and no one will notice.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 671290,
      "author_name": "anjum48",
      "author_url": "",
      "post_date": "11/12/2019 13:42:32",
      "content": "<p>In the Humpback whale competition, the LB was altered to remove a small leak:\n<a href=\"https://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664\">https://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664</a></p>\n\n<p>Maybe the same can be done here? </p>\n\n<p>It's also interesting to see how different the first half of the train data is compared to the test data for the UCF site. This has a pretty big impact on what models for this site could look like.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "670185": "Kaggle has already mentioned that they are aware of data leaks but they haven't mentioned how they will handle them. \n\nWill leaked segments of the data be removed from the private LB? I think it's important to know so we can work out what data to use in our models.",
    "670354": "This explains why some people jumped suddenly to the top of LB.\nIt has started and this is not gonna stop. LB will become useless.",
    "670552": "In general everyone should ask himself what are his goals at Kaggle?  For me it's better to spend time to becoming smarter in EDA, ML but looking for the leaks and exploit them.",
    "670884": "In my opinion, the competition data should be limited within Kaggle only with no reference to external dataset so that it is fair for every participant.",
    "671290": "In the Humpback whale competition, the LB was altered to remove a small leak:\nhttps://www.kaggle.com/c/humpback-whale-identification/discussion/81341#latest-475664\n\nMaybe the same can be done here? \n\nIt's also interesting to see how different the first half of the train data is compared to the test data for the UCF site. This has a pretty big impact on what models for this site could look like.",
    "672233": "Is it not a good sign that using the leaked data improves the LB score - this suggests that at least some or most of the leaked data is not part of the private score.",
    "680376": "I agree.The host should have restricted use of external data.This has now become a webscraping competition.✋",
    "680380": "How can you detect the person uses it or not? Someone can easily slip leaks to their submissions like a blend and no one will notice."
  },
  "source": "meta"
}