{
  "id": 274175,
  "title": "Public NB having High Scores are Useless ",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/274175",
  "author_name": "",
  "post_date": "2021-09-24T11:54:18.691123300Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "<p>To anyone who is new , you should be aware that using a public notebook which is showing you a good public LB score will not be useful at all. It will ruin your ranking in the private LB </p>\n<p>One outstanding example for this is </p>\n<p><a href=\"https://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730\" target=\"_blank\">https://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730</a></p>\n<p>The notebook replaces all the IDs which are part of private LB with 0.5 , this means your score on private LB will be 0.5 for sure . </p>\n<p>Don't use or depend on such Sham Submissions . </p>",
  "messages": [
    {
      "id": "1522633",
      "postDate": "09/24/2021 11:54:18",
      "content": "<p>To anyone who is new , you should be aware that using a public notebook which is showing you a good public LB score will not be useful at all. It will ruin your ranking in the private LB </p>\n<p>One outstanding example for this is </p>\n<p><a href=\"https://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730\" target=\"_blank\">https://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730</a></p>\n<p>The notebook replaces all the IDs which are part of private LB with 0.5 , this means your score on private LB will be 0.5 for sure . </p>\n<p>Don't use or depend on such Sham Submissions . </p>",
      "rawMarkdown": "To anyone who is new , you should be aware that using a public notebook which is showing you a good public LB score will not be useful at all. It will ruin your ranking in the private LB \n\nOne outstanding example for this is \n\nhttps://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730\n\nThe notebook replaces all the IDs which are part of private LB with 0.5 , this means your score on private LB will be 0.5 for sure . \n\nDon't use or depend on such Sham Submissions .",
      "votes": null
    },
    {
      "id": "1523600",
      "postDate": "09/25/2021 14:54:54",
      "content": "<p>yes, it is useless, and they wont get any score on pr lb</p>",
      "rawMarkdown": "yes, it is useless, and they wont get any score on pr lb",
      "votes": null
    },
    {
      "id": "1523678",
      "postDate": "09/25/2021 16:03:27",
      "content": "<p>Hello. The majority understands this very well, because this has been said more than once. And those who use them, use them knowing that, most likely, they will not take the first places (although everything can be). Just as those who have 1 in lb know. Would you like to randomly choose the numbers to take first place?</p>",
      "rawMarkdown": "Hello. The majority understands this very well, because this has been said more than once. And those who use them, use them knowing that, most likely, they will not take the first places (although everything can be). Just as those who have 1 in lb know. Would you like to randomly choose the numbers to take first place?",
      "votes": null
    },
    {
      "id": "1524320",
      "postDate": "09/26/2021 11:58:16",
      "content": "<p>Even if the private samples were classified with the used models, it would be worthless too. The notebook highly overfits public LB using an ensemble of models that accidentally achieve good scores on the public LB, but do not generalize well. The ensemble weights for averaging the probabilities are also prepared to achieve the best result on the public LB only. This cannot work well on the private dataset</p>",
      "rawMarkdown": "Even if the private samples were classified with the used models, it would be worthless too. The notebook highly overfits public LB using an ensemble of models that accidentally achieve good scores on the public LB, but do not generalize well. The ensemble weights for averaging the probabilities are also prepared to achieve the best result on the public LB only. This cannot work well on the private dataset",
      "votes": null
    },
    {
      "id": "1525103",
      "postDate": "09/27/2021 06:18:07",
      "content": "<p>It has been already mentioned clearly in the notebook</p>",
      "rawMarkdown": "It has been already mentioned clearly in the notebook",
      "votes": null
    },
    {
      "id": "1525106",
      "postDate": "09/27/2021 06:23:45",
      "content": "<p>He just added it recently. The first version didn't mention it at all.</p>",
      "rawMarkdown": "He just added it recently. The first version didn't mention it at all.",
      "votes": null
    },
    {
      "id": "1525110",
      "postDate": "09/27/2021 06:28:52",
      "content": "<p>ok             </p>",
      "rawMarkdown": "ok",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1523600,
      "author_name": "hijest",
      "author_url": "",
      "post_date": "09/25/2021 14:54:54",
      "content": "<p>yes, it is useless, and they wont get any score on pr lb</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1523678,
      "author_name": "zaakciiru",
      "author_url": "",
      "post_date": "09/25/2021 16:03:27",
      "content": "<p>Hello. The majority understands this very well, because this has been said more than once. And those who use them, use them knowing that, most likely, they will not take the first places (although everything can be). Just as those who have 1 in lb know. Would you like to randomly choose the numbers to take first place?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1524320,
      "author_name": "mikecho",
      "author_url": "",
      "post_date": "09/26/2021 11:58:16",
      "content": "<p>Even if the private samples were classified with the used models, it would be worthless too. The notebook highly overfits public LB using an ensemble of models that accidentally achieve good scores on the public LB, but do not generalize well. The ensemble weights for averaging the probabilities are also prepared to achieve the best result on the public LB only. This cannot work well on the private dataset</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1525103,
      "author_name": "swaralipibose",
      "author_url": "",
      "post_date": "09/27/2021 06:18:07",
      "content": "<p>It has been already mentioned clearly in the notebook</p>",
      "votes": null,
      "replies": [
        {
          "id": 1525106,
          "author_name": "nanguyen",
          "author_url": "",
          "post_date": "09/27/2021 06:23:45",
          "content": "<p>He just added it recently. The first version didn't mention it at all.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1525110,
          "author_name": "swaralipibose",
          "author_url": "",
          "post_date": "09/27/2021 06:28:52",
          "content": "<p>ok             </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1522633": "To anyone who is new , you should be aware that using a public notebook which is showing you a good public LB score will not be useful at all. It will ruin your ranking in the private LB \n\nOne outstanding example for this is \n\nhttps://www.kaggle.com/ammarnassanalhajali/brain-tumor-3d-blender-lb-0-730\n\nThe notebook replaces all the IDs which are part of private LB with 0.5 , this means your score on private LB will be 0.5 for sure . \n\nDon't use or depend on such Sham Submissions .",
    "1523600": "yes, it is useless, and they wont get any score on pr lb",
    "1523678": "Hello. The majority understands this very well, because this has been said more than once. And those who use them, use them knowing that, most likely, they will not take the first places (although everything can be). Just as those who have 1 in lb know. Would you like to randomly choose the numbers to take first place?",
    "1524320": "Even if the private samples were classified with the used models, it would be worthless too. The notebook highly overfits public LB using an ensemble of models that accidentally achieve good scores on the public LB, but do not generalize well. The ensemble weights for averaging the probabilities are also prepared to achieve the best result on the public LB only. This cannot work well on the private dataset",
    "1525103": "It has been already mentioned clearly in the notebook",
    "1525106": "He just added it recently. The first version didn't mention it at all.",
    "1525110": "ok"
  },
  "source": "meta"
}