{
  "id": 19521,
  "title": "Code for 19th solutions",
  "url": "/competitions/second-annual-data-science-bowl/discussion/19521",
  "author_name": "udibr",
  "post_date": "2016-03-15T01:05:25.043000",
  "votes": 7,
  "comment_count": 0,
  "views": 600,
  "content": "<p>Congratulations to the winners!</p>\n\n<p><a href=\"https://github.com/udibr/DSB2/blob/master/160306-Readme.ipynb\">My code</a> reached the 19th place but I think it has useful ideas:</p>\n\n<ul>\n<li>converted time to DC and first 2 cos and sin frequencies (5 channels)</li>\n<li>each horizontal cropped slice is feed into a CNN which predicts the volume contribution of each slice to the entire volume of the heart</li>\n<li>when predicting, the results from the same study are added up</li>\n<li>when training a special arrangement is used in which all slices from the same study appear in the same batch and the loss function sums all slices from the same study before computing loss</li>\n<li>CNN predicts both the volume and the error of the prediction and the loss is negative log likelihood of a normal distribution</li>\n</ul>",
  "messages": [
    {
      "id": 111503,
      "postDate": "2016-03-15T01:05:25.043Z",
      "content": "<p>Congratulations to the winners!</p>\n\n<p><a href=\"https://github.com/udibr/DSB2/blob/master/160306-Readme.ipynb\">My code</a> reached the 19th place but I think it has useful ideas:</p>\n\n<ul>\n<li>converted time to DC and first 2 cos and sin frequencies (5 channels)</li>\n<li>each horizontal cropped slice is feed into a CNN which predicts the volume contribution of each slice to the entire volume of the heart</li>\n<li>when predicting, the results from the same study are added up</li>\n<li>when training a special arrangement is used in which all slices from the same study appear in the same batch and the loss function sums all slices from the same study before computing loss</li>\n<li>CNN predicts both the volume and the error of the prediction and the loss is negative log likelihood of a normal distribution</li>\n</ul>",
      "rawMarkdown": "Congratulations to the winners!\r\n\r\n [My code][1] reached the 19th place but I think it has useful ideas:\r\n\r\n* converted time to DC and first 2 cos and sin frequencies (5 channels)\r\n* each horizontal cropped slice is feed into a CNN which predicts the volume contribution of each slice to the entire volume of the heart\r\n* when predicting, the results from the same study are added up\r\n* when training a special arrangement is used in which all slices from the same study appear in the same batch and the loss function sums all slices from the same study before computing loss\r\n* CNN predicts both the volume and the error of the prediction and the loss is negative log likelihood of a normal distribution\r\n\r\n\r\n  [1]: https://github.com/udibr/DSB2/blob/master/160306-Readme.ipynb",
      "votes": 7
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "111503": "Congratulations to the winners!\r\n\r\n [My code][1] reached the 19th place but I think it has useful ideas:\r\n\r\n* converted time to DC and first 2 cos and sin frequencies (5 channels)\r\n* each horizontal cropped slice is feed into a CNN which predicts the volume contribution of each slice to the entire volume of the heart\r\n* when predicting, the results from the same study are added up\r\n* when training a special arrangement is used in which all slices from the same study appear in the same batch and the loss function sums all slices from the same study before computing loss\r\n* CNN predicts both the volume and the error of the prediction and the loss is negative log likelihood of a normal distribution\r\n\r\n\r\n  [1]: https://github.com/udibr/DSB2/blob/master/160306-Readme.ipynb"
  }
}