{
  "id": 185821,
  "title": "Thinking of joining",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/185821",
  "author_name": "Louka Ewington-Pitsos",
  "post_date": "2020-09-22T08:32:07.371000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey, I'm thinking of joining this competition, the data seems interesting. I'm just wondering if there is any reason there are so few competitors though. Is there some issue with this competition that is driving people away?</p>",
  "messages": [
    {
      "id": 1022904,
      "postDate": "2020-09-22T19:59:18.130Z",
      "content": "<p>I think the size of the dataset and the need to make submissions through a notebook has stopped the usual \"Leaderboard climbing\" that often accounts for a lot of entries. This competition doesn't lend itself to five random submissions a day with a tiny change to some blending parameter.</p>\n<p>JPEG datasets have been posted, which helps with the size of the dataset.</p>\n<p>I find it nice not to be competing with a lot of \"overfitters\".</p>\n<p>This competition provides the opportunity to learn some important data scientist skills:</p>\n<ul>\n<li><p>large dataset</p></li>\n<li><p>DICOM processing - many AI tasks are related to medical imaging, so knowing how to handle DICOM is important. In the real world, the data is not packaged for you in train.csv files and JPEGs.</p></li>\n<li><p>Good chance to use TPUs for training (although you cannot use them for Inference of the \"private\" test set).</p></li>\n<li><p>Real resource constraints - like the real world.</p></li>\n<li><p>A complex multi-part problem. Not just \"PE on image\", but acute vs chronic, location of PE, and a medically related but computationally distinct mini-challenge - RV/LV ratio.</p></li>\n</ul>\n<p>Hope to see you on the leaderboard!</p>\n<p>-Rich</p>",
      "rawMarkdown": "I think the size of the dataset and the need to make submissions through a notebook has stopped the usual \"Leaderboard climbing\" that often accounts for a lot of entries. This competition doesn't lend itself to five random submissions a day with a tiny change to some blending parameter.\n\nJPEG datasets have been posted, which helps with the size of the dataset.\n\nI find it nice not to be competing with a lot of \"overfitters\".\n\nThis competition provides the opportunity to learn some important data scientist skills:\n\n- large dataset\n\n- DICOM processing - many AI tasks are related to medical imaging, so knowing how to handle DICOM is important. In the real world, the data is not packaged for you in train.csv files and JPEGs.\n\n- Good chance to use TPUs for training (although you cannot use them for Inference of the \"private\" test set).\n\n- Real resource constraints - like the real world.\n\n- A complex multi-part problem. Not just \"PE on image\", but acute vs chronic, location of PE, and a medically related but computationally distinct mini-challenge - RV/LV ratio.\n\nHope to see you on the leaderboard!\n\n-Rich",
      "votes": 4,
      "replies": [
        {
          "id": 1022981,
          "postDate": "2020-09-22T21:18:08.053Z",
          "content": "<p>Dang Richard, thanks so much for such an in-depth answer to my query.  You sell the comp very well mate, I gotta join now.</p>",
          "rawMarkdown": "Dang Richard, thanks so much for such an in-depth answer to my query.  You sell the comp very well mate, I gotta join now.",
          "votes": 1
        }
      ]
    },
    {
      "id": 1021953,
      "postDate": "2020-09-22T08:32:07.373Z",
      "content": "<p>Hey, I'm thinking of joining this competition, the data seems interesting. I'm just wondering if there is any reason there are so few competitors though. Is there some issue with this competition that is driving people away?</p>",
      "rawMarkdown": "Hey, I'm thinking of joining this competition, the data seems interesting. I'm just wondering if there is any reason there are so few competitors though. Is there some issue with this competition that is driving people away?",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1022904,
      "author_name": "quadcore/Richard Epstein",
      "author_url": "",
      "post_date": "2020-09-22T19:59:18.130000",
      "content": "<p>I think the size of the dataset and the need to make submissions through a notebook has stopped the usual \"Leaderboard climbing\" that often accounts for a lot of entries. This competition doesn't lend itself to five random submissions a day with a tiny change to some blending parameter.</p>\n<p>JPEG datasets have been posted, which helps with the size of the dataset.</p>\n<p>I find it nice not to be competing with a lot of \"overfitters\".</p>\n<p>This competition provides the opportunity to learn some important data scientist skills:</p>\n<ul>\n<li><p>large dataset</p></li>\n<li><p>DICOM processing - many AI tasks are related to medical imaging, so knowing how to handle DICOM is important. In the real world, the data is not packaged for you in train.csv files and JPEGs.</p></li>\n<li><p>Good chance to use TPUs for training (although you cannot use them for Inference of the \"private\" test set).</p></li>\n<li><p>Real resource constraints - like the real world.</p></li>\n<li><p>A complex multi-part problem. Not just \"PE on image\", but acute vs chronic, location of PE, and a medically related but computationally distinct mini-challenge - RV/LV ratio.</p></li>\n</ul>\n<p>Hope to see you on the leaderboard!</p>\n<p>-Rich</p>",
      "votes": 4,
      "replies": [
        {
          "id": 1022981,
          "author_name": "Louka Ewington-Pitsos",
          "author_url": "",
          "post_date": "2020-09-22T21:18:08.053000",
          "content": "<p>Dang Richard, thanks so much for such an in-depth answer to my query.  You sell the comp very well mate, I gotta join now.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1022904": "I think the size of the dataset and the need to make submissions through a notebook has stopped the usual \"Leaderboard climbing\" that often accounts for a lot of entries. This competition doesn't lend itself to five random submissions a day with a tiny change to some blending parameter.\n\nJPEG datasets have been posted, which helps with the size of the dataset.\n\nI find it nice not to be competing with a lot of \"overfitters\".\n\nThis competition provides the opportunity to learn some important data scientist skills:\n\n- large dataset\n\n- DICOM processing - many AI tasks are related to medical imaging, so knowing how to handle DICOM is important. In the real world, the data is not packaged for you in train.csv files and JPEGs.\n\n- Good chance to use TPUs for training (although you cannot use them for Inference of the \"private\" test set).\n\n- Real resource constraints - like the real world.\n\n- A complex multi-part problem. Not just \"PE on image\", but acute vs chronic, location of PE, and a medically related but computationally distinct mini-challenge - RV/LV ratio.\n\nHope to see you on the leaderboard!\n\n-Rich",
    "1021953": "Hey, I'm thinking of joining this competition, the data seems interesting. I'm just wondering if there is any reason there are so few competitors though. Is there some issue with this competition that is driving people away?"
  }
}