{
  "id": 301514,
  "title": "Will there be a big shakeup?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/301514",
  "author_name": "",
  "post_date": "2022-01-18T05:13:27.493629800Z",
  "votes": 7,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Will there be a big shakeup?</p>",
  "messages": [
    {
      "id": "1654048",
      "postDate": "01/18/2022 05:13:27",
      "content": "<p>Will there be a big shakeup?</p>",
      "rawMarkdown": "Will there be a big shakeup?",
      "votes": null
    },
    {
      "id": "1654057",
      "postDate": "01/18/2022 05:31:33",
      "content": "<p>Most likely, I'm definitely wondering how overfit the top public notebooks are to the public LB</p>",
      "rawMarkdown": "Most likely, I'm definitely wondering how overfit the top public notebooks are to the public LB",
      "votes": null
    },
    {
      "id": "1654064",
      "postDate": "01/18/2022 05:38:43",
      "content": "<p>i hope so.<br>\nIMO main reason is : <br>\n<strong>You may select up to 4 submissions to be used to count towards your final leaderboard score. If 4 submissions are not selected, they will be automatically chosen based on your best submission scores on the public leaderboard.</strong></p>",
      "rawMarkdown": "i hope so.\nIMO main reason is : \n**You may select up to 4 submissions to be used to count towards your final leaderboard score. If 4 submissions are not selected, they will be automatically chosen based on your best submission scores on the public leaderboard.**",
      "votes": null
    },
    {
      "id": "1654105",
      "postDate": "01/18/2022 06:44:55",
      "content": "<p>It should be.  Agree with <a href=\"https://www.kaggle.com/sheep\" target=\"_blank\">@sheep</a>.　Public LB is merely a public lb.</p>",
      "rawMarkdown": "It should be.  Agree with @sheep.　Public LB is merely a public lb.",
      "votes": null
    },
    {
      "id": "1654169",
      "postDate": "01/18/2022 08:19:26",
      "content": "<p>It depends on many factors :) (consultant answer):</p>\n<ul>\n<li>test validation strategy - Kaggle can validate public LB on all training videos but score only <br>\nevery fourth frame - NO change (or small change)</li>\n<li>a new sophisticated test set - I doubt - we have 3 different movies - they consist of sequences which represent many different situations  - NO change (or small change),</li>\n<li>looking only at the public LB result - having 4 subs, you can choose different models for final submission - I am sure experienced Kagglers will do this -  NO change (or small change)</li>\n<li>one more … something strange with f2 score validator …. now I see no correlations (strange behavior) between local and LB f2 score (but this is probably due to lack of my experience) … - I know it's almost impossible … and it looks like a joke but for me it's hard to validate the local model so there is a chance we have shakeup … (but it is simillar to third bullet) - NO change (or small change)</li>\n</ul>\n<p>We will see :) </p>",
      "rawMarkdown": "It depends on many factors :) (consultant answer):\n- test validation strategy - Kaggle can validate public LB on all training videos but score only \nevery fourth frame - NO change (or small change)\n- a new sophisticated test set - I doubt - we have 3 different movies - they consist of sequences which represent many different situations  - NO change (or small change),\n- looking only at the public LB result - having 4 subs, you can choose different models for final submission - I am sure experienced Kagglers will do this -  NO change (or small change)\n- one more ... something strange with f2 score validator .... now I see no correlations (strange behavior) between local and LB f2 score (but this is probably due to lack of my experience) ... - I know it's almost impossible ... and it looks like a joke but for me it's hard to validate the local model so there is a chance we have shakeup ... (but it is simillar to third bullet) - NO change (or small change)\n\nWe will see :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1654057,
      "author_name": "maxvandijck",
      "author_url": "",
      "post_date": "01/18/2022 05:31:33",
      "content": "<p>Most likely, I'm definitely wondering how overfit the top public notebooks are to the public LB</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1654064,
      "author_name": "mobassir",
      "author_url": "",
      "post_date": "01/18/2022 05:38:43",
      "content": "<p>i hope so.<br>\nIMO main reason is : <br>\n<strong>You may select up to 4 submissions to be used to count towards your final leaderboard score. If 4 submissions are not selected, they will be automatically chosen based on your best submission scores on the public leaderboard.</strong></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1654105,
      "author_name": "shigengtian",
      "author_url": "",
      "post_date": "01/18/2022 06:44:55",
      "content": "<p>It should be.  Agree with <a href=\"https://www.kaggle.com/sheep\" target=\"_blank\">@sheep</a>.　Public LB is merely a public lb.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1654169,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "01/18/2022 08:19:26",
      "content": "<p>It depends on many factors :) (consultant answer):</p>\n<ul>\n<li>test validation strategy - Kaggle can validate public LB on all training videos but score only <br>\nevery fourth frame - NO change (or small change)</li>\n<li>a new sophisticated test set - I doubt - we have 3 different movies - they consist of sequences which represent many different situations  - NO change (or small change),</li>\n<li>looking only at the public LB result - having 4 subs, you can choose different models for final submission - I am sure experienced Kagglers will do this -  NO change (or small change)</li>\n<li>one more … something strange with f2 score validator …. now I see no correlations (strange behavior) between local and LB f2 score (but this is probably due to lack of my experience) … - I know it's almost impossible … and it looks like a joke but for me it's hard to validate the local model so there is a chance we have shakeup … (but it is simillar to third bullet) - NO change (or small change)</li>\n</ul>\n<p>We will see :) </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1654048": "Will there be a big shakeup?",
    "1654057": "Most likely, I'm definitely wondering how overfit the top public notebooks are to the public LB",
    "1654064": "i hope so.\nIMO main reason is : \n**You may select up to 4 submissions to be used to count towards your final leaderboard score. If 4 submissions are not selected, they will be automatically chosen based on your best submission scores on the public leaderboard.**",
    "1654105": "It should be.  Agree with @sheep.　Public LB is merely a public lb.",
    "1654169": "It depends on many factors :) (consultant answer):\n- test validation strategy - Kaggle can validate public LB on all training videos but score only \nevery fourth frame - NO change (or small change)\n- a new sophisticated test set - I doubt - we have 3 different movies - they consist of sequences which represent many different situations  - NO change (or small change),\n- looking only at the public LB result - having 4 subs, you can choose different models for final submission - I am sure experienced Kagglers will do this -  NO change (or small change)\n- one more ... something strange with f2 score validator .... now I see no correlations (strange behavior) between local and LB f2 score (but this is probably due to lack of my experience) ... - I know it's almost impossible ... and it looks like a joke but for me it's hard to validate the local model so there is a chance we have shakeup ... (but it is simillar to third bullet) - NO change (or small change)\n\nWe will see :)"
  },
  "source": "meta"
}