{
  "id": 522189,
  "title": "Hyped for 0.99 solutions",
  "url": "/competitions/uspto-explainable-ai/discussion/522189",
  "author_name": "Raki",
  "post_date": "2024-07-24T21:18:56.305000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I'm really looking forward to seeing how the top teams basically solved the problem and if it translates on hidden test. My best guesses are:</p>\n<ol>\n<li>Local query computation for all ~12M rows older than 1975 then inference just with a dictionary lookup. But that would take a ton of compute!</li>\n<li>Maybe they probed and figured out construction of the test set. 125k/200k are known of course and the remaining 75k could be selected in a specific way. Which would make this task much easier, just having to overfit  the known whoosh index.</li>\n</ol>\n<p>I also think there won't be a large shakeup.</p>\n<p>What are your guesses? </p>",
  "messages": [
    {
      "id": 2934999,
      "postDate": "2024-07-24T21:18:56.307Z",
      "content": "<p>I'm really looking forward to seeing how the top teams basically solved the problem and if it translates on hidden test. My best guesses are:</p>\n<ol>\n<li>Local query computation for all ~12M rows older than 1975 then inference just with a dictionary lookup. But that would take a ton of compute!</li>\n<li>Maybe they probed and figured out construction of the test set. 125k/200k are known of course and the remaining 75k could be selected in a specific way. Which would make this task much easier, just having to overfit  the known whoosh index.</li>\n</ol>\n<p>I also think there won't be a large shakeup.</p>\n<p>What are your guesses? </p>",
      "rawMarkdown": "I'm really looking forward to seeing how the top teams basically solved the problem and if it translates on hidden test. My best guesses are:\n\n1. Local query computation for all ~12M rows older than 1975 then inference just with a dictionary lookup. But that would take a ton of compute!\n2. Maybe they probed and figured out construction of the test set. 125k/200k are known of course and the remaining 75k could be selected in a specific way. Which would make this task much easier, just having to overfit  the known whoosh index.\n\nI also think there won't be a large shakeup.\n\nWhat are your guesses? ",
      "votes": 6
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2934999": "I'm really looking forward to seeing how the top teams basically solved the problem and if it translates on hidden test. My best guesses are:\n\n1. Local query computation for all ~12M rows older than 1975 then inference just with a dictionary lookup. But that would take a ton of compute!\n2. Maybe they probed and figured out construction of the test set. 125k/200k are known of course and the remaining 75k could be selected in a specific way. Which would make this task much easier, just having to overfit  the known whoosh index.\n\nI also think there won't be a large shakeup.\n\nWhat are your guesses? "
  }
}