{
  "id": 35097,
  "title": "Forum/Kernel based solution scoring 0.87401/0.83271",
  "url": "/competitions/intel-mobileodt-cervical-cancer-screening/writeups/mike-kim-forum-kernel-based-solution-scoring-0-874",
  "author_name": "",
  "post_date": "2017-06-22T07:10:29.177Z",
  "votes": 16,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Here's a slightly modified public forum/kernel based solution that scores 0.87401/0.83271 (private/public)\nModifications include:\n1) BatchNormalization\n2) Bagging\n3) A few more layers of Conv and Dense. 4) Slightly larger input image resize.\nShould run in maybe an hour or so on AWS P2. Props to whoever made the original public leaderboard script.</p>",
  "messages": [
    {
      "id": "194900",
      "postDate": "06/22/2017 06:18:26",
      "content": "<p>Here's a slightly modified public forum/kernel based solution that scores 0.87401/0.83271 (private/public)\nModifications include:\n1) BatchNormalization\n2) Bagging\n3) A few more layers of Conv and Dense. 4) Slightly larger input image resize.\nShould run in maybe an hour or so on AWS P2. Props to whoever made the original public leaderboard script.</p>",
      "rawMarkdown": "Here's a slightly modified public forum/kernel based solution that scores 0.87401/0.83271 (private/public)\nModifications include:\n1) BatchNormalization\n2) Bagging\n3) A few more layers of Conv and Dense. 4) Slightly larger input image resize.\nShould run in maybe an hour or so on AWS P2. Props to whoever made the original public leaderboard script.",
      "votes": null
    },
    {
      "id": "194907",
      "postDate": "06/22/2017 07:08:39",
      "content": "<p>This model scores 0.83 on 2nd stage leaderboard? Using 80x80 images? Holly cow! This is better than my 2 level ensamble of 40 models using 480x360 images!</p>",
      "rawMarkdown": "This model scores 0.83 on 2nd stage leaderboard? Using 80x80 images? Holly cow! This is better than my 2 level ensamble of 40 models using 480x360 images!",
      "votes": null
    },
    {
      "id": "194908",
      "postDate": "06/22/2017 07:10:10",
      "content": "<p>Sorry for the confusion. It scores only 0.87401 on private.</p>",
      "rawMarkdown": "Sorry for the confusion. It scores only 0.87401 on private.",
      "votes": null
    },
    {
      "id": "194914",
      "postDate": "06/22/2017 07:29:38",
      "content": "<p>Still, such a simple model would get a silver medal in this competitions.</p>",
      "rawMarkdown": "Still, such a simple model would get a silver medal in this competitions.",
      "votes": null
    },
    {
      "id": "194918",
      "postDate": "06/22/2017 07:46:58",
      "content": "<p>Hey Mike, </p>\n\n<p>You nailed it, with such a simple model. :) congrats for silver medal. </p>",
      "rawMarkdown": "Hey Mike, \n\nYou nailed it, with such a simple model. :) congrats for silver medal.",
      "votes": null
    },
    {
      "id": "194929",
      "postDate": "06/22/2017 08:34:29",
      "content": "<p>This is the best solution, I have seen so far. Fast and elegant. Congratulations!!</p>",
      "rawMarkdown": "This is the best solution, I have seen so far. Fast and elegant. Congratulations!!",
      "votes": null
    },
    {
      "id": "195626",
      "postDate": "06/24/2017 08:55:40",
      "content": "<p>Hi, I have a question. Why did you use 4 fc layers instead of 2 or 3 fc layers ?</p>",
      "rawMarkdown": "Hi, I have a question. Why did you use 4 fc layers instead of 2 or 3 fc layers ?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 194907,
      "author_name": "sakvaua",
      "author_url": "",
      "post_date": "06/22/2017 07:08:39",
      "content": "<p>This model scores 0.83 on 2nd stage leaderboard? Using 80x80 images? Holly cow! This is better than my 2 level ensamble of 40 models using 480x360 images!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 194908,
      "author_name": "mikeskim",
      "author_url": "",
      "post_date": "06/22/2017 07:10:10",
      "content": "<p>Sorry for the confusion. It scores only 0.87401 on private.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 194914,
      "author_name": "sakvaua",
      "author_url": "",
      "post_date": "06/22/2017 07:29:38",
      "content": "<p>Still, such a simple model would get a silver medal in this competitions.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 194918,
      "author_name": "svesal",
      "author_url": "",
      "post_date": "06/22/2017 07:46:58",
      "content": "<p>Hey Mike, </p>\n\n<p>You nailed it, with such a simple model. :) congrats for silver medal. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 194929,
      "author_name": "skhemka",
      "author_url": "",
      "post_date": "06/22/2017 08:34:29",
      "content": "<p>This is the best solution, I have seen so far. Fast and elegant. Congratulations!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 195626,
      "author_name": "yangyu94",
      "author_url": "",
      "post_date": "06/24/2017 08:55:40",
      "content": "<p>Hi, I have a question. Why did you use 4 fc layers instead of 2 or 3 fc layers ?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "194900": "Here's a slightly modified public forum/kernel based solution that scores 0.87401/0.83271 (private/public)\nModifications include:\n1) BatchNormalization\n2) Bagging\n3) A few more layers of Conv and Dense. 4) Slightly larger input image resize.\nShould run in maybe an hour or so on AWS P2. Props to whoever made the original public leaderboard script.",
    "194907": "This model scores 0.83 on 2nd stage leaderboard? Using 80x80 images? Holly cow! This is better than my 2 level ensamble of 40 models using 480x360 images!",
    "194908": "Sorry for the confusion. It scores only 0.87401 on private.",
    "194914": "Still, such a simple model would get a silver medal in this competitions.",
    "194918": "Hey Mike, \n\nYou nailed it, with such a simple model. :) congrats for silver medal.",
    "194929": "This is the best solution, I have seen so far. Fast and elegant. Congratulations!!",
    "195626": "Hi, I have a question. Why did you use 4 fc layers instead of 2 or 3 fc layers ?"
  },
  "source": "meta"
}