{
  "id": 74659,
  "title": "Shake-up in LB? ",
  "url": "/competitions/PLAsTiCC-2018/discussion/74659",
  "author_name": "",
  "post_date": "2018-12-14T08:54:27.258263600Z",
  "votes": 1,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Since the train and test distributions are different. Lets discuss about the possible shake up we might witness in 4 days.</p>",
  "messages": [
    {
      "id": "438832",
      "postDate": "12/14/2018 08:54:27",
      "content": "<p>Since the train and test distributions are different. Lets discuss about the possible shake up we might witness in 4 days.</p>",
      "rawMarkdown": "Since the train and test distributions are different. Lets discuss about the possible shake up we might witness in 4 days.",
      "votes": null
    },
    {
      "id": "438850",
      "postDate": "12/14/2018 09:38:03",
      "content": "<p>What is relevant here is whether there is a difference between private test and public test data.  Train has nothing to do with it.</p>",
      "rawMarkdown": "What is relevant here is whether there is a difference between private test and public test data.  Train has nothing to do with it.",
      "votes": null
    },
    {
      "id": "439006",
      "postDate": "12/14/2018 15:17:54",
      "content": "<p>sorry my bad .. but to avoid shift in private LB, i usually do 2 things :\n1)  Run 10 different seeds and average results\n2) Trusting CV\nIs there any other method to prevent the downfall on private leaderboard ?</p>",
      "rawMarkdown": "sorry my bad .. but to avoid shift in private LB, i usually do 2 things :\n1)  Run 10 different seeds and average results\n2) Trusting CV\nIs there any other method to prevent the downfall on private leaderboard ?",
      "votes": null
    },
    {
      "id": "439038",
      "postDate": "12/14/2018 16:24:44",
      "content": "<p>I think what you do is fine in general.  However, here CV is way less reliable than in other competitions because of train/test difference.  There are ways specific to this competition to prevent overfitting.  We use some, but other teams seem to use more of them then us ;)  We'll disclose ours after competition end.</p>",
      "rawMarkdown": "I think what you do is fine in general.  However, here CV is way less reliable than in other competitions because of train/test difference.  There are ways specific to this competition to prevent overfitting.  We use some, but other teams seem to use more of them then us ;)  We'll disclose ours after competition end.",
      "votes": null
    },
    {
      "id": "439069",
      "postDate": "12/14/2018 17:19:51",
      "content": "<p>Thanks :) Yeah looking forward to it</p>",
      "rawMarkdown": "Thanks :) Yeah looking forward to it",
      "votes": null
    },
    {
      "id": "454837",
      "postDate": "01/12/2019 10:38:17",
      "content": "<p>0.00900</p>",
      "rawMarkdown": "0.00900",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 438850,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "12/14/2018 09:38:03",
      "content": "<p>What is relevant here is whether there is a difference between private test and public test data.  Train has nothing to do with it.</p>",
      "votes": null,
      "replies": [
        {
          "id": 439006,
          "author_name": "prashanththangavel",
          "author_url": "",
          "post_date": "12/14/2018 15:17:54",
          "content": "<p>sorry my bad .. but to avoid shift in private LB, i usually do 2 things :\n1)  Run 10 different seeds and average results\n2) Trusting CV\nIs there any other method to prevent the downfall on private leaderboard ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 439038,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "12/14/2018 16:24:44",
          "content": "<p>I think what you do is fine in general.  However, here CV is way less reliable than in other competitions because of train/test difference.  There are ways specific to this competition to prevent overfitting.  We use some, but other teams seem to use more of them then us ;)  We'll disclose ours after competition end.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 439069,
          "author_name": "prashanththangavel",
          "author_url": "",
          "post_date": "12/14/2018 17:19:51",
          "content": "<p>Thanks :) Yeah looking forward to it</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 454837,
      "author_name": "glimmung",
      "author_url": "",
      "post_date": "01/12/2019 10:38:17",
      "content": "<p>0.00900</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "438832": "Since the train and test distributions are different. Lets discuss about the possible shake up we might witness in 4 days.",
    "438850": "What is relevant here is whether there is a difference between private test and public test data.  Train has nothing to do with it.",
    "439006": "sorry my bad .. but to avoid shift in private LB, i usually do 2 things :\n1)  Run 10 different seeds and average results\n2) Trusting CV\nIs there any other method to prevent the downfall on private leaderboard ?",
    "439038": "I think what you do is fine in general.  However, here CV is way less reliable than in other competitions because of train/test difference.  There are ways specific to this competition to prevent overfitting.  We use some, but other teams seem to use more of them then us ;)  We'll disclose ours after competition end.",
    "439069": "Thanks :) Yeah looking forward to it",
    "454837": "0.00900"
  },
  "source": "meta"
}