{
  "id": 492332,
  "title": "Running experiments quickly seems to be the key to winning competitions.",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/492332",
  "author_name": "nynyny67",
  "post_date": "2024-04-09T10:07:06.831000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>First, I would like to thank the organizers and the community of the competition!<br>\nNow I'm enjoying to see great ideas of winners!<br>\nCongratulations and thanks for sharing the winning solutions.<br>\nI will try to reproduce winning solutions in the next few weeks to study. (Please share if you have any tips for reproducing winning solutions.)</p>\n<p>In addition to the solution of the winners, I want to know the way they find great solutions through great deal of trials and errors.<br>\nFor example, if it costs you 1 hour to train a model using full dataset, you may need 100 hours to try 100 different ideas. If you use 1/10 of the dataset, you can shorten the time but the observed score gets less trustworthy.</p>\n<p>So how to make the experiments quick and accurate seems to be the key to winning the competitions.<br>\nI would appreciate if you could share the strategies you utilized in this competition.</p>",
  "messages": [
    {
      "id": 2743251,
      "postDate": "2024-04-09T10:07:06.830Z",
      "content": "<p>First, I would like to thank the organizers and the community of the competition!<br>\nNow I'm enjoying to see great ideas of winners!<br>\nCongratulations and thanks for sharing the winning solutions.<br>\nI will try to reproduce winning solutions in the next few weeks to study. (Please share if you have any tips for reproducing winning solutions.)</p>\n<p>In addition to the solution of the winners, I want to know the way they find great solutions through great deal of trials and errors.<br>\nFor example, if it costs you 1 hour to train a model using full dataset, you may need 100 hours to try 100 different ideas. If you use 1/10 of the dataset, you can shorten the time but the observed score gets less trustworthy.</p>\n<p>So how to make the experiments quick and accurate seems to be the key to winning the competitions.<br>\nI would appreciate if you could share the strategies you utilized in this competition.</p>",
      "rawMarkdown": "First, I would like to thank the organizers and the community of the competition!\nNow I'm enjoying to see great ideas of winners!\nCongratulations and thanks for sharing the winning solutions.\nI will try to reproduce winning solutions in the next few weeks to study. (Please share if you have any tips for reproducing winning solutions.)\n\nIn addition to the solution of the winners, I want to know the way they find great solutions through great deal of trials and errors.\nFor example, if it costs you 1 hour to train a model using full dataset, you may need 100 hours to try 100 different ideas. If you use 1/10 of the dataset, you can shorten the time but the observed score gets less trustworthy.\n\nSo how to make the experiments quick and accurate seems to be the key to winning the competitions.\nI would appreciate if you could share the strategies you utilized in this competition.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2743251": "First, I would like to thank the organizers and the community of the competition!\nNow I'm enjoying to see great ideas of winners!\nCongratulations and thanks for sharing the winning solutions.\nI will try to reproduce winning solutions in the next few weeks to study. (Please share if you have any tips for reproducing winning solutions.)\n\nIn addition to the solution of the winners, I want to know the way they find great solutions through great deal of trials and errors.\nFor example, if it costs you 1 hour to train a model using full dataset, you may need 100 hours to try 100 different ideas. If you use 1/10 of the dataset, you can shorten the time but the observed score gets less trustworthy.\n\nSo how to make the experiments quick and accurate seems to be the key to winning the competitions.\nI would appreciate if you could share the strategies you utilized in this competition."
  }
}