{
  "id": 428326,
  "title": "suggestion for 100k solution writeup prize",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/428326",
  "author_name": "hengck23",
  "post_date": "2023-08-01T03:14:07.178000",
  "votes": 19,
  "comment_count": 0,
  "views": 0,
  "content": "<p>please see: <a href=\"https://www.kaggle.com/discussions/general/427114#2362386\" target=\"_blank\">https://www.kaggle.com/discussions/general/427114#2362386</a></p>\n<p>As many of you know, there is a 100k solution writeup prize by kaggle.<br>\nUsually when there is shakeup, we want to know why.</p>\n<p>For example in this competition, there are some doubts:</p>\n<ul>\n<li>dilate verus non dilate</li>\n<li>which dataset for training? is only dataset1 sufficient and why?</li>\n</ul>\n<p>most competition writeup just focus on solution, but not many focus on analysis and ablation experiments (removing components of solutions to find out cause of improvement)</p>\n<p>maybe one can just take the top solutions and do more experiments and do a writeup. maybe that can win solution writeup prize.</p>\n<p>we want more experiments, graphs and visulisation ….</p>\n<hr>\n<p>e.g.:<br>\nif i have time, i will do a writeup on how to estimate private dataset uncertainy and upper bound results (e.g. via probing and experiements).<br>\ni would write about the risk (shakeup) of each solution component.<br>\nIn kaggle, it is not about getting best results only, but also predictable ones.</p>",
  "messages": [
    {
      "id": 2368152,
      "postDate": "2023-08-01T03:14:07.177Z",
      "content": "<p>please see: <a href=\"https://www.kaggle.com/discussions/general/427114#2362386\" target=\"_blank\">https://www.kaggle.com/discussions/general/427114#2362386</a></p>\n<p>As many of you know, there is a 100k solution writeup prize by kaggle.<br>\nUsually when there is shakeup, we want to know why.</p>\n<p>For example in this competition, there are some doubts:</p>\n<ul>\n<li>dilate verus non dilate</li>\n<li>which dataset for training? is only dataset1 sufficient and why?</li>\n</ul>\n<p>most competition writeup just focus on solution, but not many focus on analysis and ablation experiments (removing components of solutions to find out cause of improvement)</p>\n<p>maybe one can just take the top solutions and do more experiments and do a writeup. maybe that can win solution writeup prize.</p>\n<p>we want more experiments, graphs and visulisation ….</p>\n<hr>\n<p>e.g.:<br>\nif i have time, i will do a writeup on how to estimate private dataset uncertainy and upper bound results (e.g. via probing and experiements).<br>\ni would write about the risk (shakeup) of each solution component.<br>\nIn kaggle, it is not about getting best results only, but also predictable ones.</p>",
      "rawMarkdown": "please see: https://www.kaggle.com/discussions/general/427114#2362386\n\nAs many of you know, there is a 100k solution writeup prize by kaggle.\nUsually when there is shakeup, we want to know why.\n\nFor example in this competition, there are some doubts:\n- dilate verus non dilate\n- which dataset for training? is only dataset1 sufficient and why?\n\nmost competition writeup just focus on solution, but not many focus on analysis and ablation experiments (removing components of solutions to find out cause of improvement)\n\nmaybe one can just take the top solutions and do more experiments and do a writeup. maybe that can win solution writeup prize.\n\nwe want more experiments, graphs and visulisation ....\n\n---\n\ne.g.:\nif i have time, i will do a writeup on how to estimate private dataset uncertainy and upper bound results (e.g. via probing and experiements).\ni would write about the risk (shakeup) of each solution component.\nIn kaggle, it is not about getting best results only, but also predictable ones.\n\n\n",
      "votes": 19
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2368152": "please see: https://www.kaggle.com/discussions/general/427114#2362386\n\nAs many of you know, there is a 100k solution writeup prize by kaggle.\nUsually when there is shakeup, we want to know why.\n\nFor example in this competition, there are some doubts:\n- dilate verus non dilate\n- which dataset for training? is only dataset1 sufficient and why?\n\nmost competition writeup just focus on solution, but not many focus on analysis and ablation experiments (removing components of solutions to find out cause of improvement)\n\nmaybe one can just take the top solutions and do more experiments and do a writeup. maybe that can win solution writeup prize.\n\nwe want more experiments, graphs and visulisation ....\n\n---\n\ne.g.:\nif i have time, i will do a writeup on how to estimate private dataset uncertainy and upper bound results (e.g. via probing and experiements).\ni would write about the risk (shakeup) of each solution component.\nIn kaggle, it is not about getting best results only, but also predictable ones.\n\n\n"
  }
}