{
  "id": 56218,
  "title": "Serenity now",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/56218",
  "author_name": "",
  "post_date": "2018-05-07T22:27:15.496210600Z",
  "votes": 6,
  "comment_count": 8,
  "views": 0,
  "content": "<p>There is legally wrong, and just plain wrong. One can be the latter without being the former. Still, has anyone actually looked into the infamous 0.9811 script (not linking it)? Is there anyone who can guarantee that the result is reliable, even if we account that it was done on a full train dataset?</p>\n\n<p>I don't think we need to be so territorial about LB positions. It seems to me there is a fair chance that script will be overfitting like many others. It just so happens to have the highest LB score because it is using most data and the largest validation dataset (though I would not necessarily consider 5M data points as reliable). Let's see what happens when the dust settles. My feeling is that the only upward mobility from that script will happen to those who didn't use it.</p>\n\n<p><img src=\"https://thumbs.gfycat.com/BasicDiligentIrishredandwhitesetter-size_restricted.gif\" alt=\"enter image description here\"></p>\n\n<p>Here is an abbreviated <a href=\"https://www.youtube.com/watch?v=Ow_9MglZrhs\"><strong>video</strong></a> if the GIF is not helping your relaxation.</p>",
  "messages": [
    {
      "id": "324693",
      "postDate": "05/07/2018 22:27:15",
      "content": "<p>There is legally wrong, and just plain wrong. One can be the latter without being the former. Still, has anyone actually looked into the infamous 0.9811 script (not linking it)? Is there anyone who can guarantee that the result is reliable, even if we account that it was done on a full train dataset?</p>\n\n<p>I don't think we need to be so territorial about LB positions. It seems to me there is a fair chance that script will be overfitting like many others. It just so happens to have the highest LB score because it is using most data and the largest validation dataset (though I would not necessarily consider 5M data points as reliable). Let's see what happens when the dust settles. My feeling is that the only upward mobility from that script will happen to those who didn't use it.</p>\n\n<p><img src=\"https://thumbs.gfycat.com/BasicDiligentIrishredandwhitesetter-size_restricted.gif\" alt=\"enter image description here\"></p>\n\n<p>Here is an abbreviated <a href=\"https://www.youtube.com/watch?v=Ow_9MglZrhs\"><strong>video</strong></a> if the GIF is not helping your relaxation.</p>",
      "rawMarkdown": "There is legally wrong, and just plain wrong. One can be the latter without being the former. Still, has anyone actually looked into the infamous 0.9811 script (not linking it)? Is there anyone who can guarantee that the result is reliable, even if we account that it was done on a full train dataset?\n\nI don't think we need to be so territorial about LB positions. It seems to me there is a fair chance that script will be overfitting like many others. It just so happens to have the highest LB score because it is using most data and the largest validation dataset (though I would not necessarily consider 5M data points as reliable). Let's see what happens when the dust settles. My feeling is that the only upward mobility from that script will happen to those who didn't use it.\n\n![enter image description here][1]\n\nHere is an abbreviated [__video__](https://www.youtube.com/watch?v=Ow_9MglZrhs) if the GIF is not helping your relaxation.\n\n  [1]: https://thumbs.gfycat.com/BasicDiligentIrishredandwhitesetter-size_restricted.gif \"Serenity now\"",
      "votes": null
    },
    {
      "id": "324717",
      "postDate": "05/07/2018 22:48:46",
      "content": "<p>Hope so. We won't use it, we don't even trust it.</p>",
      "rawMarkdown": "Hope so. We won't use it, we don't even trust it.",
      "votes": null
    },
    {
      "id": "324899",
      "postDate": "05/08/2018 01:25:55",
      "content": "<p>nice GIF, Tilii. Emotion well captured! </p>",
      "rawMarkdown": "nice GIF, Tilii. Emotion well captured!",
      "votes": null
    },
    {
      "id": "325143",
      "postDate": "05/08/2018 07:00:10",
      "content": "<p>Serenity now..Insanity later!!</p>",
      "rawMarkdown": "Serenity now..Insanity later!!",
      "votes": null
    },
    {
      "id": "325440",
      "postDate": "05/08/2018 12:20:30",
      "content": "<p>So, what is the private LB score of that script?</p>",
      "rawMarkdown": "So, what is the private LB score of that script?",
      "votes": null
    },
    {
      "id": "325454",
      "postDate": "05/08/2018 12:38:46",
      "content": "<p>0.9811175 pub lb 0.9819167 private lb</p>",
      "rawMarkdown": "0.9811175 pub lb 0.9819167 private lb",
      "votes": null
    },
    {
      "id": "325528",
      "postDate": "05/08/2018 13:55:17",
      "content": "<p>Thanks.</p>",
      "rawMarkdown": "Thanks.",
      "votes": null
    },
    {
      "id": "327122",
      "postDate": "05/10/2018 20:58:53",
      "content": "<p>@Ee Kin, that sounds more like the public leaderboard score. Are you sure it's the private leaderboard score?</p>",
      "rawMarkdown": "Ee Kin, that sounds more like the public leaderboard score. Are you sure it's the private leaderboard score?",
      "votes": null
    },
    {
      "id": "327403",
      "postDate": "05/11/2018 13:01:43",
      "content": "<p>rectified thanks 👌🏻</p>",
      "rawMarkdown": "rectified thanks 👌🏻",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 324717,
      "author_name": "laevatein",
      "author_url": "",
      "post_date": "05/07/2018 22:48:46",
      "content": "<p>Hope so. We won't use it, we don't even trust it.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 324899,
      "author_name": "dongxu027",
      "author_url": "",
      "post_date": "05/08/2018 01:25:55",
      "content": "<p>nice GIF, Tilii. Emotion well captured! </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 325143,
      "author_name": "abhilashawasthi",
      "author_url": "",
      "post_date": "05/08/2018 07:00:10",
      "content": "<p>Serenity now..Insanity later!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 325440,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "05/08/2018 12:20:30",
      "content": "<p>So, what is the private LB score of that script?</p>",
      "votes": null,
      "replies": [
        {
          "id": 325454,
          "author_name": "dicksonchin93",
          "author_url": "",
          "post_date": "05/08/2018 12:38:46",
          "content": "<p>0.9811175 pub lb 0.9819167 private lb</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 325528,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "05/08/2018 13:55:17",
          "content": "<p>Thanks.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 327122,
          "author_name": "araksstepanyan",
          "author_url": "",
          "post_date": "05/10/2018 20:58:53",
          "content": "<p>@Ee Kin, that sounds more like the public leaderboard score. Are you sure it's the private leaderboard score?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 327403,
          "author_name": "dicksonchin93",
          "author_url": "",
          "post_date": "05/11/2018 13:01:43",
          "content": "<p>rectified thanks 👌🏻</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "324693": "There is legally wrong, and just plain wrong. One can be the latter without being the former. Still, has anyone actually looked into the infamous 0.9811 script (not linking it)? Is there anyone who can guarantee that the result is reliable, even if we account that it was done on a full train dataset?\n\nI don't think we need to be so territorial about LB positions. It seems to me there is a fair chance that script will be overfitting like many others. It just so happens to have the highest LB score because it is using most data and the largest validation dataset (though I would not necessarily consider 5M data points as reliable). Let's see what happens when the dust settles. My feeling is that the only upward mobility from that script will happen to those who didn't use it.\n\n![enter image description here][1]\n\nHere is an abbreviated [__video__](https://www.youtube.com/watch?v=Ow_9MglZrhs) if the GIF is not helping your relaxation.\n\n  [1]: https://thumbs.gfycat.com/BasicDiligentIrishredandwhitesetter-size_restricted.gif \"Serenity now\"",
    "324717": "Hope so. We won't use it, we don't even trust it.",
    "324899": "nice GIF, Tilii. Emotion well captured!",
    "325143": "Serenity now..Insanity later!!",
    "325440": "So, what is the private LB score of that script?",
    "325454": "0.9811175 pub lb 0.9819167 private lb",
    "325528": "Thanks.",
    "327122": "Ee Kin, that sounds more like the public leaderboard score. Are you sure it's the private leaderboard score?",
    "327403": "rectified thanks 👌🏻"
  },
  "source": "meta"
}