{
  "id": 107959,
  "title": "Competition next year? and feedback",
  "url": "/competitions/aptos2019-blindness-detection/discussion/107959",
  "author_name": "ilovescience",
  "post_date": "2019-09-08T06:37:02.393000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Typically, when a competition's name includes the year, it means that there will be annual competitions (ex: Data Science Bowl, Google Open Images, iNaturalist, Freesound, etc.). Are we lead to believe that there will be another competition next year?</p>\n\n<p>If so, ** that's great! **</p>\n\n<p>but...</p>\n\n<p>There are several problems that should be avoided for next year:\n1. It should not be a KO competition <em>unless</em> the kernel infrastructure becomes stable. This was a major problem for this competition.\n2. Computing resources should be more available. 3/4 way through the competition, Kaggle severely limited our access to GPUs. In a future competition, it would be nice to, say, apply for GCP credits for the competition.\n3. The training set and test set should be better constructed such that they are similar and to prevent such a large shakeup. The image size and class distributions should be similar, allowing us to better evaluate and develop better models.</p>\n\n<p>All in all, I enjoyed this competition, with an intriguing problem to solve. I would be excited to work on a newer iteration of this competition, as long as the above problems are solved.</p>",
  "messages": [
    {
      "id": 621013,
      "postDate": "2019-09-08T06:37:02.393Z",
      "content": "<p>Typically, when a competition's name includes the year, it means that there will be annual competitions (ex: Data Science Bowl, Google Open Images, iNaturalist, Freesound, etc.). Are we lead to believe that there will be another competition next year?</p>\n\n<p>If so, ** that's great! **</p>\n\n<p>but...</p>\n\n<p>There are several problems that should be avoided for next year:\n1. It should not be a KO competition <em>unless</em> the kernel infrastructure becomes stable. This was a major problem for this competition.\n2. Computing resources should be more available. 3/4 way through the competition, Kaggle severely limited our access to GPUs. In a future competition, it would be nice to, say, apply for GCP credits for the competition.\n3. The training set and test set should be better constructed such that they are similar and to prevent such a large shakeup. The image size and class distributions should be similar, allowing us to better evaluate and develop better models.</p>\n\n<p>All in all, I enjoyed this competition, with an intriguing problem to solve. I would be excited to work on a newer iteration of this competition, as long as the above problems are solved.</p>",
      "rawMarkdown": "Typically, when a competition's name includes the year, it means that there will be annual competitions (ex: Data Science Bowl, Google Open Images, iNaturalist, Freesound, etc.). Are we lead to believe that there will be another competition next year?\n\nIf so, ** that's great! **\n\nbut...\n \nThere are several problems that should be avoided for next year:\n1. It should not be a KO competition _unless_ the kernel infrastructure becomes stable. This was a major problem for this competition.\n2. Computing resources should be more available. 3/4 way through the competition, Kaggle severely limited our access to GPUs. In a future competition, it would be nice to, say, apply for GCP credits for the competition.\n3. The training set and test set should be better constructed such that they are similar and to prevent such a large shakeup. The image size and class distributions should be similar, allowing us to better evaluate and develop better models.\n\nAll in all, I enjoyed this competition, with an intriguing problem to solve. I would be excited to work on a newer iteration of this competition, as long as the above problems are solved.",
      "votes": 6
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "621013": "Typically, when a competition's name includes the year, it means that there will be annual competitions (ex: Data Science Bowl, Google Open Images, iNaturalist, Freesound, etc.). Are we lead to believe that there will be another competition next year?\n\nIf so, ** that's great! **\n\nbut...\n \nThere are several problems that should be avoided for next year:\n1. It should not be a KO competition _unless_ the kernel infrastructure becomes stable. This was a major problem for this competition.\n2. Computing resources should be more available. 3/4 way through the competition, Kaggle severely limited our access to GPUs. In a future competition, it would be nice to, say, apply for GCP credits for the competition.\n3. The training set and test set should be better constructed such that they are similar and to prevent such a large shakeup. The image size and class distributions should be similar, allowing us to better evaluate and develop better models.\n\nAll in all, I enjoyed this competition, with an intriguing problem to solve. I would be excited to work on a newer iteration of this competition, as long as the above problems are solved."
  }
}