{
  "id": 227741,
  "title": "congratz to XU",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/227741",
  "author_name": "",
  "post_date": "2021-03-22T03:08:01.941942300Z",
  "votes": 12,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi all,</p>\n<p>First of all, thanks to the host and kaggle team for giving us such interesting question to tackle, and all the winners for providing such insightful solutions. I have learned a lot from all the solutions posted after competition (especial 1st team's segmentation + classification ensemble, very impressive)</p>\n<p>However, I am still curious to Xu <a href=\"https://www.kaggle.com/wowfattie\" target=\"_blank\">@wowfattie</a> as I follow him through several competitions and personally learned a lot from his sharing. As I remembered, before the competition deadline, Xu was about 120 and the result was super impressive. I am wondering what would be this impressive cv strategy, if Xu is willing to share a bit.</p>\n<p>Thanks a lot. 👍</p>\n<p>Best, </p>",
  "messages": [
    {
      "id": "1247757",
      "postDate": "03/22/2021 03:08:01",
      "content": "<p>Hi all,</p>\n<p>First of all, thanks to the host and kaggle team for giving us such interesting question to tackle, and all the winners for providing such insightful solutions. I have learned a lot from all the solutions posted after competition (especial 1st team's segmentation + classification ensemble, very impressive)</p>\n<p>However, I am still curious to Xu <a href=\"https://www.kaggle.com/wowfattie\" target=\"_blank\">@wowfattie</a> as I follow him through several competitions and personally learned a lot from his sharing. As I remembered, before the competition deadline, Xu was about 120 and the result was super impressive. I am wondering what would be this impressive cv strategy, if Xu is willing to share a bit.</p>\n<p>Thanks a lot. 👍</p>\n<p>Best, </p>",
      "rawMarkdown": "Hi all,\n\nFirst of all, thanks to the host and kaggle team for giving us such interesting question to tackle, and all the winners for providing such insightful solutions. I have learned a lot from all the solutions posted after competition (especial 1st team's segmentation + classification ensemble, very impressive)\n\nHowever, I am still curious to Xu @wowfattie as I follow him through several competitions and personally learned a lot from his sharing. As I remembered, before the competition deadline, Xu was about 120 and the result was super impressive. I am wondering what would be this impressive cv strategy, if Xu is willing to share a bit.\n\nThanks a lot. 👍\n\nBest,",
      "votes": null
    },
    {
      "id": "1254032",
      "postDate": "03/27/2021 09:31:13",
      "content": "<p>In this comp I doubt if any approach that works well on local CV can't score in the top 100 of the public leaderboard. My guess is he uses the trick to omit the public test dataset and he realized the big correlation between local CV and the public test dataset. In this case, he doesnt need to submit to public LB :)</p>",
      "rawMarkdown": "In this comp I doubt if any approach that works well on local CV can't score in the top 100 of the public leaderboard. My guess is he uses the trick to omit the public test dataset and he realized the big correlation between local CV and the public test dataset. In this case, he doesnt need to submit to public LB :)",
      "votes": null
    },
    {
      "id": "1254157",
      "postDate": "03/27/2021 11:37:34",
      "content": "<p>I guess he could squeeze in some extra models during inference time by not having to inference on public images in private set</p>",
      "rawMarkdown": "I guess he could squeeze in some extra models during inference time by not having to inference on public images in private set",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1254032,
      "author_name": "moewie94",
      "author_url": "",
      "post_date": "03/27/2021 09:31:13",
      "content": "<p>In this comp I doubt if any approach that works well on local CV can't score in the top 100 of the public leaderboard. My guess is he uses the trick to omit the public test dataset and he realized the big correlation between local CV and the public test dataset. In this case, he doesnt need to submit to public LB :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1254157,
          "author_name": "raddar",
          "author_url": "",
          "post_date": "03/27/2021 11:37:34",
          "content": "<p>I guess he could squeeze in some extra models during inference time by not having to inference on public images in private set</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1247757": "Hi all,\n\nFirst of all, thanks to the host and kaggle team for giving us such interesting question to tackle, and all the winners for providing such insightful solutions. I have learned a lot from all the solutions posted after competition (especial 1st team's segmentation + classification ensemble, very impressive)\n\nHowever, I am still curious to Xu @wowfattie as I follow him through several competitions and personally learned a lot from his sharing. As I remembered, before the competition deadline, Xu was about 120 and the result was super impressive. I am wondering what would be this impressive cv strategy, if Xu is willing to share a bit.\n\nThanks a lot. 👍\n\nBest,",
    "1254032": "In this comp I doubt if any approach that works well on local CV can't score in the top 100 of the public leaderboard. My guess is he uses the trick to omit the public test dataset and he realized the big correlation between local CV and the public test dataset. In this case, he doesnt need to submit to public LB :)",
    "1254157": "I guess he could squeeze in some extra models during inference time by not having to inference on public images in private set"
  },
  "source": "meta"
}