{
  "id": 20482,
  "title": "Getting started with nolearn/lasagne",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/20482",
  "author_name": "",
  "post_date": "2016-04-27T13:31:19.773Z",
  "votes": 6,
  "comment_count": 2,
  "views": 978,
  "content": "<p>During our internal hackathon at Otto Group, we played a little bit with the State Farm dataset. We thought it would be nice to share our efforts with the community.</p>\n\n<p>Check out our repo at <a href=\"https://github.com/ottogroup/statefarm\">https://github.com/ottogroup/statefarm</a>.</p>\n\n<p>We use nolearn/lasagne and get a local CV score of &lt; 0.9 when splitting on drivers. We're sure you can improve on that ;)</p>",
  "messages": [
    {
      "id": "117123",
      "postDate": "04/27/2016 13:31:19",
      "content": "<p>During our internal hackathon at Otto Group, we played a little bit with the State Farm dataset. We thought it would be nice to share our efforts with the community.</p>\n\n<p>Check out our repo at <a href=\"https://github.com/ottogroup/statefarm\">https://github.com/ottogroup/statefarm</a>.</p>\n\n<p>We use nolearn/lasagne and get a local CV score of &lt; 0.9 when splitting on drivers. We're sure you can improve on that ;)</p>",
      "rawMarkdown": "During our internal hackathon at Otto Group, we played a little bit with the State Farm dataset. We thought it would be nice to share our efforts with the community.\r\n\r\nCheck out our repo at https://github.com/ottogroup/statefarm.\r\n\r\nWe use nolearn/lasagne and get a local CV score of < 0.9 when splitting on drivers. We're sure you can improve on that ;)",
      "votes": null
    },
    {
      "id": "117175",
      "postDate": "04/27/2016 17:39:33",
      "content": "<p>Did right cropping the images help ?</p>",
      "rawMarkdown": "Did right cropping the images help ?",
      "votes": null
    },
    {
      "id": "117196",
      "postDate": "04/27/2016 20:32:03",
      "content": "<p>It was mostly to get square images and the right side seemed to contain the most relevant information. But maybe the net could also learn the correct label from looking at the notepad on the woman's leg :)</p>",
      "rawMarkdown": "It was mostly to get square images and the right side seemed to contain the most relevant information. But maybe the net could also learn the correct label from looking at the notepad on the woman's leg :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 117175,
      "author_name": "jonomon",
      "author_url": "",
      "post_date": "04/27/2016 17:39:33",
      "content": "<p>Did right cropping the images help ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 117196,
      "author_name": "benjaminbossan",
      "author_url": "",
      "post_date": "04/27/2016 20:32:03",
      "content": "<p>It was mostly to get square images and the right side seemed to contain the most relevant information. But maybe the net could also learn the correct label from looking at the notepad on the woman's leg :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "117123": "During our internal hackathon at Otto Group, we played a little bit with the State Farm dataset. We thought it would be nice to share our efforts with the community.\r\n\r\nCheck out our repo at https://github.com/ottogroup/statefarm.\r\n\r\nWe use nolearn/lasagne and get a local CV score of < 0.9 when splitting on drivers. We're sure you can improve on that ;)",
    "117175": "Did right cropping the images help ?",
    "117196": "It was mostly to get square images and the right side seemed to contain the most relevant information. But maybe the net could also learn the correct label from looking at the notepad on the woman's leg :)"
  },
  "source": "meta"
}