{
  "id": 11367,
  "title": "Script for visualization of training data",
  "url": "/competitions/datasciencebowl/discussion/11367",
  "author_name": "",
  "post_date": "2014-12-27T01:29:51.490Z",
  "votes": 46,
  "comment_count": 6,
  "views": 2736,
  "content": "<p>I wanted to share some scripts I've been working on to help with visualizing the training data.</p>\n<p>The first compiles all the images from each class into a mosaic image, and the second creates a bubble chart of these mosaics where related classes are grouped together.</p>\n<p>Since the full chart is extremely&nbsp;large, I used&nbsp;<a href=\"http://polymaps.org/ex/\">Polymaps</a>&nbsp;to get&nbsp;a pan/zoomable web interface.</p>\n<p>Source and instructions are available on&nbsp;<a href=\"https://github.com/ebenolson/kaggle-ndsb-visualization/tree/master\">Github</a>, give it a try and&nbsp;if you have ideas for improvements please let me know or submit a pull request. I do plan to add a mode to show model predictions side by side with the training data.</p>\n<p>Some screenshots:</p>\n<p>Single mosaic:</p>\n<p><img src=\"http://i.imgur.com/C5nFvzM.png\" alt=\"A single mosaic\" width=\"334\" height=\"500\"></p>\n<p>Zoomed out completely:</p>\n<p><img src=\"http://i.imgur.com/VDO7Ilp.png\" alt=\"Zoomed out\" width=\"1246\" height=\"781\"></p>\n<p>Zooming in on one group:</p>\n<p><img src=\"http://i.imgur.com/brizfzs.png\" alt=\"Zoomed in on one group\" width=\"1436\" height=\"774\"></p>",
  "messages": [
    {
      "id": "61921",
      "postDate": "12/27/2014 01:29:51",
      "content": "<p>I wanted to share some scripts I've been working on to help with visualizing the training data.</p>\n<p>The first compiles all the images from each class into a mosaic image, and the second creates a bubble chart of these mosaics where related classes are grouped together.</p>\n<p>Since the full chart is extremely&nbsp;large, I used&nbsp;<a href=\"http://polymaps.org/ex/\">Polymaps</a>&nbsp;to get&nbsp;a pan/zoomable web interface.</p>\n<p>Source and instructions are available on&nbsp;<a href=\"https://github.com/ebenolson/kaggle-ndsb-visualization/tree/master\">Github</a>, give it a try and&nbsp;if you have ideas for improvements please let me know or submit a pull request. I do plan to add a mode to show model predictions side by side with the training data.</p>\n<p>Some screenshots:</p>\n<p>Single mosaic:</p>\n<p><img src=\"http://i.imgur.com/C5nFvzM.png\" alt=\"A single mosaic\" width=\"334\" height=\"500\"></p>\n<p>Zoomed out completely:</p>\n<p><img src=\"http://i.imgur.com/VDO7Ilp.png\" alt=\"Zoomed out\" width=\"1246\" height=\"781\"></p>\n<p>Zooming in on one group:</p>\n<p><img src=\"http://i.imgur.com/brizfzs.png\" alt=\"Zoomed in on one group\" width=\"1436\" height=\"774\"></p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61933",
      "postDate": "12/27/2014 15:15:22",
      "content": "<p>Thanks for sharing! This is really useful stuff.&nbsp;</p>\n<p>I've uploaded the generated mosaics here, for everyone's convenience:</p>\n<p>http://npow.github.io/plankton/viewer/index.html</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61939",
      "postDate": "12/27/2014 16:17:45",
      "content": "<p>Very nice, thanks to both of you!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61940",
      "postDate": "12/27/2014 16:44:49",
      "content": "<p>Thanks guys, Really Useful.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61961",
      "postDate": "12/27/2014 23:32:22",
      "content": "<p>So cool. Thanks a lot, guys</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61966",
      "postDate": "12/28/2014 00:56:12",
      "content": "<p>Thanks, looks interesting!</p>\n<p>There's one thing I'm quite curious about - does the aspect ratios of images leak information (I suspect they do, not sure about the severity)?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61971",
      "postDate": "12/28/2014 02:59:45",
      "content": "<p>Thanks. Good way to visualize.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 61933,
      "author_name": "nissanpow",
      "author_url": "",
      "post_date": "12/27/2014 15:15:22",
      "content": "<p>Thanks for sharing! This is really useful stuff.&nbsp;</p>\n<p>I've uploaded the generated mosaics here, for everyone's convenience:</p>\n<p>http://npow.github.io/plankton/viewer/index.html</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61939,
      "author_name": "sedielem",
      "author_url": "",
      "post_date": "12/27/2014 16:17:45",
      "content": "<p>Very nice, thanks to both of you!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61940,
      "author_name": "",
      "author_url": "",
      "post_date": "12/27/2014 16:44:49",
      "content": "<p>Thanks guys, Really Useful.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61961,
      "author_name": "tudinhnguyen",
      "author_url": "",
      "post_date": "12/27/2014 23:32:22",
      "content": "<p>So cool. Thanks a lot, guys</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61966,
      "author_name": "ziyaowei",
      "author_url": "",
      "post_date": "12/28/2014 00:56:12",
      "content": "<p>Thanks, looks interesting!</p>\n<p>There's one thing I'm quite curious about - does the aspect ratios of images leak information (I suspect they do, not sure about the severity)?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61971,
      "author_name": "liangpang",
      "author_url": "",
      "post_date": "12/28/2014 02:59:45",
      "content": "<p>Thanks. Good way to visualize.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "61921": "",
    "61933": "",
    "61939": "",
    "61940": "",
    "61961": "",
    "61966": "",
    "61971": ""
  },
  "source": "meta"
}