{
  "id": 174410,
  "title": " i was completely at a loss...",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/174410",
  "author_name": "shiba",
  "post_date": "2020-08-13T12:46:42.599000",
  "votes": 0,
  "comment_count": 8,
  "views": 0,
  "content": "<p>more TTA make CV increase, LB decrease!!<br>\nensemble models make CV increase, LB almost no change<br>\nwhat should i do ?…Even these simple methods cant improve the my LB…<br>\nMy code is pure pytorch.</p>",
  "messages": [
    {
      "id": 969094,
      "postDate": "2020-08-13T13:01:32.197Z",
      "content": "<p>Again, in CV we trust…</p>",
      "rawMarkdown": "Again, in CV we trust...",
      "votes": 3,
      "replies": [
        {
          "id": 969107,
          "postDate": "2020-08-13T13:09:12.667Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 969108,
          "postDate": "2020-08-13T13:10:20.083Z",
          "content": "<p>😹 Hope soo…</p>",
          "rawMarkdown": "😹 Hope soo..."
        }
      ]
    },
    {
      "id": 970261,
      "postDate": "2020-08-14T09:56:01.400Z",
      "content": "<p>This has to be one of the hardest competitions I've worked on from a mental standpoint. It's very difficult to find the motivation to work on this problem without the gratification of climbing up the public leaderboard.</p>\n<p>I just keep telling myself to trust my CV and hopefully, everything will be ok in a few days 🤞️</p>",
      "rawMarkdown": "This has to be one of the hardest competitions I've worked on from a mental standpoint. It's very difficult to find the motivation to work on this problem without the gratification of climbing up the public leaderboard.\n\nI just keep telling myself to trust my CV and hopefully, everything will be ok in a few days 🤞️",
      "votes": 1
    },
    {
      "id": 969293,
      "postDate": "2020-08-13T15:41:41.823Z",
      "content": "<p>This is one of the challenges of the current competition: should you trust your local CV or the LB score? Betting on the right strategy can make a very big difference on the private leader board. </p>",
      "rawMarkdown": "This is one of the challenges of the current competition: should you trust your local CV or the LB score? Betting on the right strategy can make a very big difference on the private leader board. ",
      "votes": 1
    },
    {
      "id": 969303,
      "postDate": "2020-08-13T15:49:19.003Z",
      "content": "<p>As first action, i would find CV which are proportional to Public LB score instead of submitting and checking Public LB score</p>",
      "rawMarkdown": "As first action, i would find CV which are proportional to Public LB score instead of submitting and checking Public LB score",
      "votes": 2
    },
    {
      "id": 969633,
      "postDate": "2020-08-13T20:11:50.367Z",
      "content": "<p>do you have a link to your code?  Surprising that TTA would make things decrease.  Are you sure you are doing the averaging properly in the TTA step? I mean collecting each iteration of TTA and then dividing by the number of TTA etc?</p>",
      "rawMarkdown": "do you have a link to your code?  Surprising that TTA would make things decrease.  Are you sure you are doing the averaging properly in the TTA step? I mean collecting each iteration of TTA and then dividing by the number of TTA etc?"
    },
    {
      "id": 969071,
      "postDate": "2020-08-13T12:46:42.600Z",
      "content": "<p>more TTA make CV increase, LB decrease!!<br>\nensemble models make CV increase, LB almost no change<br>\nwhat should i do ?…Even these simple methods cant improve the my LB…<br>\nMy code is pure pytorch.</p>",
      "rawMarkdown": "more TTA make CV increase, LB decrease!!\nensemble models make CV increase, LB almost no change\nwhat should i do ?...Even these simple methods cant improve the my LB...\nMy code is pure pytorch."
    },
    {
      "id": 969300,
      "postDate": "2020-08-13T15:47:38.397Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 969094,
      "author_name": "Quan",
      "author_url": "",
      "post_date": "2020-08-13T13:01:32.197000",
      "content": "<p>Again, in CV we trust…</p>",
      "votes": 3,
      "replies": [
        {
          "id": 969107,
          "author_name": "",
          "author_url": "",
          "post_date": "2020-08-13T13:09:12.667000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 969108,
          "author_name": "shiba",
          "author_url": "",
          "post_date": "2020-08-13T13:10:20.083000",
          "content": "<p>😹 Hope soo…</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 970261,
      "author_name": "datasaurus",
      "author_url": "",
      "post_date": "2020-08-14T09:56:01.400000",
      "content": "<p>This has to be one of the hardest competitions I've worked on from a mental standpoint. It's very difficult to find the motivation to work on this problem without the gratification of climbing up the public leaderboard.</p>\n<p>I just keep telling myself to trust my CV and hopefully, everything will be ok in a few days 🤞️</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 969293,
      "author_name": "Alexey Pronin",
      "author_url": "",
      "post_date": "2020-08-13T15:41:41.823000",
      "content": "<p>This is one of the challenges of the current competition: should you trust your local CV or the LB score? Betting on the right strategy can make a very big difference on the private leader board. </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 969303,
      "author_name": "kaggler",
      "author_url": "",
      "post_date": "2020-08-13T15:49:19.003000",
      "content": "<p>As first action, i would find CV which are proportional to Public LB score instead of submitting and checking Public LB score</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 969633,
      "author_name": "Signal",
      "author_url": "",
      "post_date": "2020-08-13T20:11:50.367000",
      "content": "<p>do you have a link to your code?  Surprising that TTA would make things decrease.  Are you sure you are doing the averaging properly in the TTA step? I mean collecting each iteration of TTA and then dividing by the number of TTA etc?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 969300,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-08-13T15:47:38.397000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "969094": "Again, in CV we trust...",
    "970261": "This has to be one of the hardest competitions I've worked on from a mental standpoint. It's very difficult to find the motivation to work on this problem without the gratification of climbing up the public leaderboard.\n\nI just keep telling myself to trust my CV and hopefully, everything will be ok in a few days 🤞️",
    "969293": "This is one of the challenges of the current competition: should you trust your local CV or the LB score? Betting on the right strategy can make a very big difference on the private leader board. ",
    "969303": "As first action, i would find CV which are proportional to Public LB score instead of submitting and checking Public LB score",
    "969633": "do you have a link to your code?  Surprising that TTA would make things decrease.  Are you sure you are doing the averaging properly in the TTA step? I mean collecting each iteration of TTA and then dividing by the number of TTA etc?",
    "969071": "more TTA make CV increase, LB decrease!!\nensemble models make CV increase, LB almost no change\nwhat should i do ?...Even these simple methods cant improve the my LB...\nMy code is pure pytorch.",
    "969300": ""
  }
}