{
  "id": 307209,
  "title": "Which 4 submissions will you choose?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/307209",
  "author_name": "",
  "post_date": "2022-02-13T08:09:54.153018400Z",
  "votes": 6,
  "comment_count": 8,
  "views": 0,
  "content": "<p>I think the shake would be a Jigsaw shake.<br>\nHow do you choose to win this lottery game?</p>",
  "messages": [
    {
      "id": "1687873",
      "postDate": "02/13/2022 08:09:54",
      "content": "<p>I think the shake would be a Jigsaw shake.<br>\nHow do you choose to win this lottery game?</p>",
      "rawMarkdown": "I think the shake would be a Jigsaw shake.\nHow do you choose to win this lottery game?",
      "votes": null
    },
    {
      "id": "1687959",
      "postDate": "02/13/2022 09:09:27",
      "content": "<ol>\n<li>highest LB score<br>\n2 - 4: top 3 CV score</li>\n</ol>",
      "rawMarkdown": "1. highest LB score\n2 - 4: top 3 CV score",
      "votes": null
    },
    {
      "id": "1687961",
      "postDate": "02/13/2022 09:11:10",
      "content": "<p>Given that we are definitely overfitting to the public lb, we'll probably choose the 3 submissions that perform best internally + the best public lb one.</p>",
      "rawMarkdown": "Given that we are definitely overfitting to the public lb, we'll probably choose the 3 submissions that perform best internally + the best public lb one.",
      "votes": null
    },
    {
      "id": "1687995",
      "postDate": "02/13/2022 09:48:15",
      "content": "<p>I don't think that the shake up would be that big.  4 selections definitely would reduce the shake up.</p>",
      "rawMarkdown": "I don't think that the shake up would be that big.  4 selections definitely would reduce the shake up.",
      "votes": null
    },
    {
      "id": "1688262",
      "postDate": "02/13/2022 14:43:01",
      "content": "<p>My first two submissions will be the ones which generalized better over the 3 clips. If the model generalize good the f2 score being calculated separately on those 3 clips valid data should give similar scores. For me, this type of models which respect this condition present more confidence than models where CVprediction scores over one clip is larger but not showing the same consistency over the other 2. <br>\nThe other 2 submissions will be chosen with an weighted average between a consistency factor described latter, the maximum CV score over the best scoring clip and the public score.</p>",
      "rawMarkdown": "My first two submissions will be the ones which generalized better over the 3 clips. If the model generalize good the f2 score being calculated separately on those 3 clips valid data should give similar scores. For me, this type of models which respect this condition present more confidence than models where CVprediction scores over one clip is larger but not showing the same consistency over the other 2. \nThe other 2 submissions will be chosen with an weighted average between a consistency factor described latter, the maximum CV score over the best scoring clip and the public score.",
      "votes": null
    },
    {
      "id": "1688308",
      "postDate": "02/13/2022 15:14:36",
      "content": "<p>plan</p>\n<ol>\n<li>highest LB score</li>\n<li>2nd high LB score</li>\n<li>3rd high LB score</li>\n<li>tf model with highest LB score</li>\n</ol>\n<p>Oh ..I think i might ask my ancestor and buddha tomorrow.</p>",
      "rawMarkdown": "plan\n1.  highest LB score\n2. 2nd high LB score\n3. 3rd high LB score\n4. tf model with highest LB score\n\nOh ..I think i might ask my ancestor and buddha tomorrow.",
      "votes": null
    },
    {
      "id": "1688317",
      "postDate": "02/13/2022 15:20:27",
      "content": "<p>1 LB, 2 Logical things, 1 lottery</p>",
      "rawMarkdown": "1 LB, 2 Logical things, 1 lottery",
      "votes": null
    },
    {
      "id": "1688521",
      "postDate": "02/13/2022 17:18:38",
      "content": "<p>4 submissions in 0.08 increments from conf = 0.05 to 0.29<br>\nOf course, the model is what I judged to be \"the most robust\".</p>",
      "rawMarkdown": "4 submissions in 0.08 increments from conf = 0.05 to 0.29\nOf course, the model is what I judged to be \"the most robust\".",
      "votes": null
    },
    {
      "id": "1688885",
      "postDate": "02/13/2022 21:56:01",
      "content": "<p>I am so bad at this part. In the Pet Finder comp I (we) had 27 submissions that would have given a better score than the ones we actually chose. Seven of which would have given us a medal!</p>\n<p>I'm sure 4 choices will help though.</p>\n<ol>\n<li>best LB</li>\n<li>best LB ensemble</li>\n<li>best larger model (Y5m)</li>\n<li>chicken bones</li>\n</ol>",
      "rawMarkdown": "I am so bad at this part. In the Pet Finder comp I (we) had 27 submissions that would have given a better score than the ones we actually chose. Seven of which would have given us a medal!\n\nI'm sure 4 choices will help though.\n\n1. best LB\n2. best LB ensemble\n3. best larger model (Y5m)\n4. chicken bones",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1687959,
      "author_name": "snaker",
      "author_url": "",
      "post_date": "02/13/2022 09:09:27",
      "content": "<ol>\n<li>highest LB score<br>\n2 - 4: top 3 CV score</li>\n</ol>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1687961,
      "author_name": "dauriel",
      "author_url": "",
      "post_date": "02/13/2022 09:11:10",
      "content": "<p>Given that we are definitely overfitting to the public lb, we'll probably choose the 3 submissions that perform best internally + the best public lb one.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1687995,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/13/2022 09:48:15",
      "content": "<p>I don't think that the shake up would be that big.  4 selections definitely would reduce the shake up.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1688262,
      "author_name": "vladvdv",
      "author_url": "",
      "post_date": "02/13/2022 14:43:01",
      "content": "<p>My first two submissions will be the ones which generalized better over the 3 clips. If the model generalize good the f2 score being calculated separately on those 3 clips valid data should give similar scores. For me, this type of models which respect this condition present more confidence than models where CVprediction scores over one clip is larger but not showing the same consistency over the other 2. <br>\nThe other 2 submissions will be chosen with an weighted average between a consistency factor described latter, the maximum CV score over the best scoring clip and the public score.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1688308,
      "author_name": "atom1231",
      "author_url": "",
      "post_date": "02/13/2022 15:14:36",
      "content": "<p>plan</p>\n<ol>\n<li>highest LB score</li>\n<li>2nd high LB score</li>\n<li>3rd high LB score</li>\n<li>tf model with highest LB score</li>\n</ol>\n<p>Oh ..I think i might ask my ancestor and buddha tomorrow.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1688317,
      "author_name": "deepkim",
      "author_url": "",
      "post_date": "02/13/2022 15:20:27",
      "content": "<p>1 LB, 2 Logical things, 1 lottery</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1688521,
      "author_name": "syurenuko",
      "author_url": "",
      "post_date": "02/13/2022 17:18:38",
      "content": "<p>4 submissions in 0.08 increments from conf = 0.05 to 0.29<br>\nOf course, the model is what I judged to be \"the most robust\".</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1688885,
      "author_name": "mutantspore",
      "author_url": "",
      "post_date": "02/13/2022 21:56:01",
      "content": "<p>I am so bad at this part. In the Pet Finder comp I (we) had 27 submissions that would have given a better score than the ones we actually chose. Seven of which would have given us a medal!</p>\n<p>I'm sure 4 choices will help though.</p>\n<ol>\n<li>best LB</li>\n<li>best LB ensemble</li>\n<li>best larger model (Y5m)</li>\n<li>chicken bones</li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1687873": "I think the shake would be a Jigsaw shake.\nHow do you choose to win this lottery game?",
    "1687959": "1. highest LB score\n2 - 4: top 3 CV score",
    "1687961": "Given that we are definitely overfitting to the public lb, we'll probably choose the 3 submissions that perform best internally + the best public lb one.",
    "1687995": "I don't think that the shake up would be that big.  4 selections definitely would reduce the shake up.",
    "1688262": "My first two submissions will be the ones which generalized better over the 3 clips. If the model generalize good the f2 score being calculated separately on those 3 clips valid data should give similar scores. For me, this type of models which respect this condition present more confidence than models where CVprediction scores over one clip is larger but not showing the same consistency over the other 2. \nThe other 2 submissions will be chosen with an weighted average between a consistency factor described latter, the maximum CV score over the best scoring clip and the public score.",
    "1688308": "plan\n1.  highest LB score\n2. 2nd high LB score\n3. 3rd high LB score\n4. tf model with highest LB score\n\nOh ..I think i might ask my ancestor and buddha tomorrow.",
    "1688317": "1 LB, 2 Logical things, 1 lottery",
    "1688521": "4 submissions in 0.08 increments from conf = 0.05 to 0.29\nOf course, the model is what I judged to be \"the most robust\".",
    "1688885": "I am so bad at this part. In the Pet Finder comp I (we) had 27 submissions that would have given a better score than the ones we actually chose. Seven of which would have given us a medal!\n\nI'm sure 4 choices will help though.\n\n1. best LB\n2. best LB ensemble\n3. best larger model (Y5m)\n4. chicken bones"
  },
  "source": "meta"
}