{
  "id": 107917,
  "title": "Congratulations to all!",
  "url": "/competitions/aptos2019-blindness-detection/discussion/107917",
  "author_name": "",
  "post_date": "2019-09-08T00:11:20.859941200Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p><strong>Congratulations to all the participants!</strong> I hope you were not grealty affected by the shake-up. </p>\n\n<p>Congrats to the winners <a href=\"/wowfattie\">@wowfattie</a> <a href=\"/dxchen\">@dxchen</a> <a href=\"/winninghealth\">@winninghealth</a> <a href=\"/lanjunyelan\">@lanjunyelan</a> <a href=\"/artemis874\">@artemis874</a> <a href=\"/firenero\">@firenero</a> <a href=\"/aispiriants\">@aispiriants</a> ! I look forward to hearing about your solutions and learn from them.</p>\n\n<p>I learned a lot from this competition and I hope you guys too. </p>\n\n<p>And I just want to say, even though there were many challenges, from GPU limits to kernel problems, <strong>we perservered</strong>! </p>",
  "messages": [
    {
      "id": "620752",
      "postDate": "09/08/2019 00:11:20",
      "content": "<p><strong>Congratulations to all the participants!</strong> I hope you were not grealty affected by the shake-up. </p>\n\n<p>Congrats to the winners <a href=\"/wowfattie\">@wowfattie</a> <a href=\"/dxchen\">@dxchen</a> <a href=\"/winninghealth\">@winninghealth</a> <a href=\"/lanjunyelan\">@lanjunyelan</a> <a href=\"/artemis874\">@artemis874</a> <a href=\"/firenero\">@firenero</a> <a href=\"/aispiriants\">@aispiriants</a> ! I look forward to hearing about your solutions and learn from them.</p>\n\n<p>I learned a lot from this competition and I hope you guys too. </p>\n\n<p>And I just want to say, even though there were many challenges, from GPU limits to kernel problems, <strong>we perservered</strong>! </p>",
      "rawMarkdown": "**Congratulations to all the participants!** I hope you were not grealty affected by the shake-up. \n\nCongrats to the winners @wowfattie @dxchen @winninghealth @lanjunyelan @artemis874 @firenero @aispiriants ! I look forward to hearing about your solutions and learn from them.\n\nI learned a lot from this competition and I hope you guys too. \n\nAnd I just want to say, even though there were many challenges, from GPU limits to kernel problems, **we perservered**!",
      "votes": null
    },
    {
      "id": "620759",
      "postDate": "09/08/2019 00:17:56",
      "content": "<p>I came nowhere near getting a medal (I didn't expect to), but I was very surprised that although I had a very big drop in standing, my private score is much higher than my public score was.  .733 vs .889.   Higher scores on the private set seems to be the case for a lot of people, which surprises me. I expected it to be the other way around. </p>",
      "rawMarkdown": "I came nowhere near getting a medal (I didn't expect to), but I was very surprised that although I had a very big drop in standing, my private score is much higher than my public score was.  .733 vs .889.   Higher scores on the private set seems to be the case for a lot of people, which surprises me. I expected it to be the other way around.",
      "votes": null
    },
    {
      "id": "620770",
      "postDate": "09/08/2019 00:28:25",
      "content": "<p>Somebody correct me if I am wrong, but doesn't this indicate that the distribution of the private set is similar to the public test set? Because we chose models that did well on the public LB, these models will output a distribution similar to the public test set, and this distribution must therefore be very similar to that of the private test set for us to get such high scores.</p>",
      "rawMarkdown": "Somebody correct me if I am wrong, but doesn't this indicate that the distribution of the private set is similar to the public test set? Because we chose models that did well on the public LB, these models will output a distribution similar to the public test set, and this distribution must therefore be very similar to that of the private test set for us to get such high scores.",
      "votes": null
    },
    {
      "id": "620785",
      "postDate": "09/08/2019 00:54:37",
      "content": "<p>Distribution is similar to train set, local CV matches private LB perfectly. </p>",
      "rawMarkdown": "Distribution is similar to train set, local CV matches private LB perfectly.",
      "votes": null
    },
    {
      "id": "620791",
      "postDate": "09/08/2019 00:58:46",
      "content": "<p>Looking at couple of my kernels, I agree.</p>",
      "rawMarkdown": "Looking at couple of my kernels, I agree.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 620759,
      "author_name": "chrisfs",
      "author_url": "",
      "post_date": "09/08/2019 00:17:56",
      "content": "<p>I came nowhere near getting a medal (I didn't expect to), but I was very surprised that although I had a very big drop in standing, my private score is much higher than my public score was.  .733 vs .889.   Higher scores on the private set seems to be the case for a lot of people, which surprises me. I expected it to be the other way around. </p>",
      "votes": null,
      "replies": [
        {
          "id": 620770,
          "author_name": "tanlikesmath",
          "author_url": "",
          "post_date": "09/08/2019 00:28:25",
          "content": "<p>Somebody correct me if I am wrong, but doesn't this indicate that the distribution of the private set is similar to the public test set? Because we chose models that did well on the public LB, these models will output a distribution similar to the public test set, and this distribution must therefore be very similar to that of the private test set for us to get such high scores.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 620785,
          "author_name": "naivelamb",
          "author_url": "",
          "post_date": "09/08/2019 00:54:37",
          "content": "<p>Distribution is similar to train set, local CV matches private LB perfectly. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 620791,
          "author_name": "tanlikesmath",
          "author_url": "",
          "post_date": "09/08/2019 00:58:46",
          "content": "<p>Looking at couple of my kernels, I agree.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "620752": "**Congratulations to all the participants!** I hope you were not grealty affected by the shake-up. \n\nCongrats to the winners @wowfattie @dxchen @winninghealth @lanjunyelan @artemis874 @firenero @aispiriants ! I look forward to hearing about your solutions and learn from them.\n\nI learned a lot from this competition and I hope you guys too. \n\nAnd I just want to say, even though there were many challenges, from GPU limits to kernel problems, **we perservered**!",
    "620759": "I came nowhere near getting a medal (I didn't expect to), but I was very surprised that although I had a very big drop in standing, my private score is much higher than my public score was.  .733 vs .889.   Higher scores on the private set seems to be the case for a lot of people, which surprises me. I expected it to be the other way around.",
    "620770": "Somebody correct me if I am wrong, but doesn't this indicate that the distribution of the private set is similar to the public test set? Because we chose models that did well on the public LB, these models will output a distribution similar to the public test set, and this distribution must therefore be very similar to that of the private test set for us to get such high scores.",
    "620785": "Distribution is similar to train set, local CV matches private LB perfectly.",
    "620791": "Looking at couple of my kernels, I agree."
  },
  "source": "meta"
}