{
  "id": 73751,
  "title": "Is there  anyone's score suddenly get smaller today? ",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/73751",
  "author_name": "",
  "post_date": "2018-12-05T11:21:29.831800300Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>My score was 0.529 earlier , but now it is  0.513.And the scores I submitted earlier are smaller now.I used to adjust the model according to this score.But now it seems that I am busy in vain, and everything has to start over again.  Is this Kaggle bug? Anyone  know why？</p>",
  "messages": [
    {
      "id": "433702",
      "postDate": "12/05/2018 11:21:29",
      "content": "<p>My score was 0.529 earlier , but now it is  0.513.And the scores I submitted earlier are smaller now.I used to adjust the model according to this score.But now it seems that I am busy in vain, and everything has to start over again.  Is this Kaggle bug? Anyone  know why？</p>",
      "rawMarkdown": "My score was 0.529 earlier , but now it is  0.513.And the scores I submitted earlier are smaller now.I used to adjust the model according to this score.But now it seems that I am busy in vain, and everything has to start over again.  Is this Kaggle bug? Anyone  know why？",
      "votes": null
    },
    {
      "id": "433707",
      "postDate": "12/05/2018 11:35:12",
      "content": "<p>I would think this has to do with the data leak. The images that were involved were probably removed from the scoring, hence different scores</p>",
      "rawMarkdown": "I would think this has to do with the data leak. The images that were involved were probably removed from the scoring, hence different scores",
      "votes": null
    },
    {
      "id": "433714",
      "postDate": "12/05/2018 11:43:13",
      "content": "<p>In that case, there may be some trouble. How to judge whether the model has improved? Just using validation set? But my model validation set macro F1 score can reach 0.91.  It seems that we can't judge whether the model is improved only by validation set.               </p>",
      "rawMarkdown": "In that case, there may be some trouble. How to judge whether the model has improved? Just using validation set? But my model validation set macro F1 score can reach 0.91.  It seems that we can't judge whether the model is improved only by validation set.",
      "votes": null
    },
    {
      "id": "434255",
      "postDate": "12/06/2018 05:32:07",
      "content": "<p>I also  have  this problem, what does  the office worker handle this problem? Remove some image from test dataset or other?</p>",
      "rawMarkdown": "I also  have  this problem, what does  the office worker handle this problem? Remove some image from test dataset or other?",
      "votes": null
    },
    {
      "id": "434515",
      "postDate": "12/06/2018 14:28:35",
      "content": "<p><a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395</a>\nyou can see  how the office worker  handle this problem at  the above link. </p>",
      "rawMarkdown": "https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395\nyou can see  how the office worker  handle this problem at  the above link.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 433707,
      "author_name": "moshel",
      "author_url": "",
      "post_date": "12/05/2018 11:35:12",
      "content": "<p>I would think this has to do with the data leak. The images that were involved were probably removed from the scoring, hence different scores</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 433714,
      "author_name": "zymale",
      "author_url": "",
      "post_date": "12/05/2018 11:43:13",
      "content": "<p>In that case, there may be some trouble. How to judge whether the model has improved? Just using validation set? But my model validation set macro F1 score can reach 0.91.  It seems that we can't judge whether the model is improved only by validation set.               </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 434255,
      "author_name": "wchzhhxj",
      "author_url": "",
      "post_date": "12/06/2018 05:32:07",
      "content": "<p>I also  have  this problem, what does  the office worker handle this problem? Remove some image from test dataset or other?</p>",
      "votes": null,
      "replies": [
        {
          "id": 434515,
          "author_name": "zymale",
          "author_url": "",
          "post_date": "12/06/2018 14:28:35",
          "content": "<p><a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395</a>\nyou can see  how the office worker  handle this problem at  the above link. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "433702": "My score was 0.529 earlier , but now it is  0.513.And the scores I submitted earlier are smaller now.I used to adjust the model according to this score.But now it seems that I am busy in vain, and everything has to start over again.  Is this Kaggle bug? Anyone  know why？",
    "433707": "I would think this has to do with the data leak. The images that were involved were probably removed from the scoring, hence different scores",
    "433714": "In that case, there may be some trouble. How to judge whether the model has improved? Just using validation set? But my model validation set macro F1 score can reach 0.91.  It seems that we can't judge whether the model is improved only by validation set.",
    "434255": "I also  have  this problem, what does  the office worker handle this problem? Remove some image from test dataset or other?",
    "434515": "https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/73395\nyou can see  how the office worker  handle this problem at  the above link."
  },
  "source": "meta"
}