{
  "id": 176484,
  "title": "1868th place in 1st competition! ",
  "url": "/competitions/siim-isic-melanoma-classification/writeups/srikanthpotukuchi-1868th-place-in-1st-competition",
  "author_name": "",
  "post_date": "2020-08-21T23:19:47.391701Z",
  "votes": -12,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I joined the competition very late, just 10 days ago. So, I forked a notebook: <a href=\"https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus\" target=\"_blank\">https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus</a> </p>\n<p>So, what did I do differently? I tried changing epochs, not using coarse drop out, removing duplicates and finally I used a trick that gave a 0.007 improvement. The trick was to take max of past 2 predictions.  </p>",
  "messages": [
    {
      "id": "980844",
      "postDate": "08/21/2020 23:19:47",
      "content": "<p>I joined the competition very late, just 10 days ago. So, I forked a notebook: <a href=\"https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus\" target=\"_blank\">https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus</a> </p>\n<p>So, what did I do differently? I tried changing epochs, not using coarse drop out, removing duplicates and finally I used a trick that gave a 0.007 improvement. The trick was to take max of past 2 predictions.  </p>",
      "rawMarkdown": "I joined the competition very late, just 10 days ago. So, I forked a notebook: https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus \n\nSo, what did I do differently? I tried changing epochs, not using coarse drop out, removing duplicates and finally I used a trick that gave a 0.007 improvement. The trick was to take max of past 2 predictions.",
      "votes": null
    },
    {
      "id": "989296",
      "postDate": "08/28/2020 17:40:15",
      "content": "<p>\"Kill them with success and bury them with a smile.\" :)</p>",
      "rawMarkdown": "\"Kill them with success and bury them with a smile.\" :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 989296,
      "author_name": "srikanthpotukuchi",
      "author_url": "",
      "post_date": "08/28/2020 17:40:15",
      "content": "<p>\"Kill them with success and bury them with a smile.\" :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "980844": "I joined the competition very late, just 10 days ago. So, I forked a notebook: https://www.kaggle.com/niteshx2/full-pipeline-dual-input-cnn-model-with-tpus \n\nSo, what did I do differently? I tried changing epochs, not using coarse drop out, removing duplicates and finally I used a trick that gave a 0.007 improvement. The trick was to take max of past 2 predictions.",
    "989296": "\"Kill them with success and bury them with a smile.\" :)"
  },
  "source": "meta"
}