{
  "id": 128417,
  "title": "30th Place Write Up",
  "content": "<p>I'd like to share what I learned a lot from this competition with you.</p>\n\n<h1>Our Approach(public 622th → private 30th)</h1>\n\n<ul>\n<li>Trust CV and LB.</li>\n<li>LGB with about 500 features.</li>\n<li>Some features were dropped by adversarial score(0.65 or so).</li>\n<li>Train by regression and then optimize with nelder-mead.</li>\n</ul>\n\n<h1>Late Submission</h1>\n\n<p>I realized adversarial validation was useless, and \"Trust CV\" is the best approach here after some late submissions.\nAlso, if I used about 2k features, maybe I could get prize as well as gold medal and make my team mates Grandmasters.\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2Fd985dcd0bb2969058ad6d94d859c32b4%2Fimage%20(3\" alt=\"\">.png?generation=1580450134030141&amp;alt=media)\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F310636e14f81322a1f7407b657cfe993%2Fimage%20(4\" alt=\"\">.png?generation=1580450231030144&amp;alt=media)</p>\n\n<h1>How to get gold medal</h1>\n\n<ul>\n<li>Generate about 30k features.</li>\n<li>Use same condition as evaluation for validation.</li>\n<li>Trust only CV(after confirmed if we can trust LB).\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F378c588fc6bf87651928ffdf2c346a70%2FQWK_random_truncate.png?generation=1580452115832570&amp;alt=media\" alt=\"\"></li>\n<li>Use a lot of features till CV saturated.</li>\n<li>Don't worry about Adversarial Validation after all.</li>\n</ul>",
  "comments": [
    {
      "id": 750130,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-02-19T06:45:15.047000",
      "content": "<p>amazing iam trying to understand code u deserve to be in first place tq u</p>",
      "votes": 15,
      "replies": []
    },
    {
      "id": 738085,
      "author_name": "Charadrius",
      "author_url": "",
      "post_date": "2020-02-06T05:45:23.027000",
      "content": "<p>Thank you for sharing !　It was helpful for me.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 736935,
      "author_name": "Subbu Vidyasekar",
      "author_url": "",
      "post_date": "2020-02-04T18:09:22.800000",
      "content": "<p>the simple and realistic post... kudo's to u</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 734417,
      "author_name": "toshi_k",
      "author_url": "",
      "post_date": "2020-02-01T12:55:58.847000",
      "content": "<p>Thank you for sharing ! I think some features which increased adversarial score were not important by chance.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 733764,
      "author_name": "dasmehdixtr",
      "author_url": "",
      "post_date": "2020-01-31T13:53:37.840000",
      "content": "<p>Thanks for share! 💯 </p>",
      "votes": 0,
      "replies": []
    }
  ],
  "comment_count": 5,
  "messages": [
    {
      "id": 733454,
      "postDate": "2020-01-31T06:33:21.357Z",
      "content": "<p>I'd like to share what I learned a lot from this competition with you.</p>\n\n<h1>Our Approach(public 622th → private 30th)</h1>\n\n<ul>\n<li>Trust CV and LB.</li>\n<li>LGB with about 500 features.</li>\n<li>Some features were dropped by adversarial score(0.65 or so).</li>\n<li>Train by regression and then optimize with nelder-mead.</li>\n</ul>\n\n<h1>Late Submission</h1>\n\n<p>I realized adversarial validation was useless, and \"Trust CV\" is the best approach here after some late submissions.\nAlso, if I used about 2k features, maybe I could get prize as well as gold medal and make my team mates Grandmasters.\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2Fd985dcd0bb2969058ad6d94d859c32b4%2Fimage%20(3\" alt=\"\">.png?generation=1580450134030141&amp;alt=media)\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F310636e14f81322a1f7407b657cfe993%2Fimage%20(4\" alt=\"\">.png?generation=1580450231030144&amp;alt=media)</p>\n\n<h1>How to get gold medal</h1>\n\n<ul>\n<li>Generate about 30k features.</li>\n<li>Use same condition as evaluation for validation.</li>\n<li>Trust only CV(after confirmed if we can trust LB).\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F378c588fc6bf87651928ffdf2c346a70%2FQWK_random_truncate.png?generation=1580452115832570&amp;alt=media\" alt=\"\"></li>\n<li>Use a lot of features till CV saturated.</li>\n<li>Don't worry about Adversarial Validation after all.</li>\n</ul>",
      "rawMarkdown": "I'd like to share what I learned a lot from this competition with you.\n\n# Our Approach(public 622th → private 30th)\n- Trust CV and LB.\n- LGB with about 500 features.\n- Some features were dropped by adversarial score(0.65 or so).\n- Train by regression and then optimize with nelder-mead.\n\n# Late Submission\nI realized adversarial validation was useless, and \"Trust CV\" is the best approach here after some late submissions.\nAlso, if I used about 2k features, maybe I could get prize as well as gold medal and make my team mates Grandmasters.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2Fd985dcd0bb2969058ad6d94d859c32b4%2Fimage%20(3).png?generation=1580450134030141&amp;alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F310636e14f81322a1f7407b657cfe993%2Fimage%20(4).png?generation=1580450231030144&amp;alt=media)\n\n# How to get gold medal\n- Generate about 30k features.\n- Use same condition as evaluation for validation.\n- Trust only CV(after confirmed if we can trust LB).\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F317344%2F378c588fc6bf87651928ffdf2c346a70%2FQWK_random_truncate.png?generation=1580452115832570&amp;alt=media)\n- Use a lot of features till CV saturated.\n- Don't worry about Adversarial Validation after all.",
      "votes": 36
    },
    {
      "id": 750130,
      "postDate": "2020-02-19T06:45:15.047Z",
      "content": "<p>amazing iam trying to understand code u deserve to be in first place tq u</p>",
      "rawMarkdown": "amazing iam trying to understand code u deserve to be in first place tq u",
      "votes": 15
    },
    {
      "id": 738085,
      "postDate": "2020-02-06T05:45:23.027Z",
      "content": "<p>Thank you for sharing !　It was helpful for me.</p>",
      "rawMarkdown": "Thank you for sharing !　It was helpful for me.\n"
    },
    {
      "id": 736935,
      "postDate": "2020-02-04T18:09:22.800Z",
      "content": "<p>the simple and realistic post... kudo's to u</p>",
      "rawMarkdown": "the simple and realistic post... kudo's to u\n"
    },
    {
      "id": 734417,
      "postDate": "2020-02-01T12:55:58.847Z",
      "content": "<p>Thank you for sharing ! I think some features which increased adversarial score were not important by chance.</p>",
      "rawMarkdown": "Thank you for sharing ! I think some features which increased adversarial score were not important by chance."
    },
    {
      "id": 733764,
      "postDate": "2020-01-31T13:53:37.840Z",
      "content": "<p>Thanks for share! 💯 </p>",
      "rawMarkdown": "Thanks for share! 💯 "
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-23T23:51:07Z",
  "competition_id": 16531,
  "competition_slug": "data-science-bowl-2019",
  "team_id": 3855227,
  "solution_url": "/competitions/data-science-bowl-2019/writeups/gm-30th-place-write-up"
}