{
  "id": 412275,
  "title": "How high a score indicates your model is working -at all- ?",
  "url": "/competitions/birdclef-2023/discussion/412275",
  "author_name": "",
  "post_date": "2023-05-23T03:28:57.259195900Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>After a bunch of tweaking - I finally got a submission to finish before timing out or throwing an error.  Considering I started the contest just a few days ago - I'm happy to have gotten this far.  </p>\n<p>My first attempt has scored .74 - will admit I was hoping for better.</p>\n<p>I'm starting from-scratch - using levit_256 trained on spectrograms of 5-second segments of the training data.  I'm aware this probably isn't the most effective approach.</p>\n<p>Here's the question - is a score of .74 any better than random guesses + a little luck? </p>\n<p>Any thoughts appreciated.</p>\n<p>-Rich</p>",
  "messages": [
    {
      "id": "2270144",
      "postDate": "05/23/2023 03:28:57",
      "content": "<p>After a bunch of tweaking - I finally got a submission to finish before timing out or throwing an error.  Considering I started the contest just a few days ago - I'm happy to have gotten this far.  </p>\n<p>My first attempt has scored .74 - will admit I was hoping for better.</p>\n<p>I'm starting from-scratch - using levit_256 trained on spectrograms of 5-second segments of the training data.  I'm aware this probably isn't the most effective approach.</p>\n<p>Here's the question - is a score of .74 any better than random guesses + a little luck? </p>\n<p>Any thoughts appreciated.</p>\n<p>-Rich</p>",
      "rawMarkdown": "After a bunch of tweaking - I finally got a submission to finish before timing out or throwing an error.  Considering I started the contest just a few days ago - I'm happy to have gotten this far.  \n\nMy first attempt has scored .74 - will admit I was hoping for better.\n\nI'm starting from-scratch - using levit_256 trained on spectrograms of 5-second segments of the training data.  I'm aware this probably isn't the most effective approach.\n\nHere's the question - is a score of .74 any better than random guesses + a little luck? \n\nAny thoughts appreciated.\n\n-Rich",
      "votes": null
    },
    {
      "id": "2270149",
      "postDate": "05/23/2023 03:40:51",
      "content": "<p>Hey rich,<br>\nRandom guess would be 0.71, as is the sample submission , reason for that lies in the way this metric is made.<br>\nyour approach isn't the most optimal for sure but it can work, my current LB is following the same approach but with a few tweaks, good luck as there is little time left but you can for sure achieve a better score.</p>",
      "rawMarkdown": "Hey rich,\nRandom guess would be 0.71, as is the sample submission , reason for that lies in the way this metric is made.\nyour approach isn't the most optimal for sure but it can work, my current LB is following the same approach but with a few tweaks, good luck as there is little time left but you can for sure achieve a better score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2270149,
      "author_name": "janmpia",
      "author_url": "",
      "post_date": "05/23/2023 03:40:51",
      "content": "<p>Hey rich,<br>\nRandom guess would be 0.71, as is the sample submission , reason for that lies in the way this metric is made.<br>\nyour approach isn't the most optimal for sure but it can work, my current LB is following the same approach but with a few tweaks, good luck as there is little time left but you can for sure achieve a better score.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2270144": "After a bunch of tweaking - I finally got a submission to finish before timing out or throwing an error.  Considering I started the contest just a few days ago - I'm happy to have gotten this far.  \n\nMy first attempt has scored .74 - will admit I was hoping for better.\n\nI'm starting from-scratch - using levit_256 trained on spectrograms of 5-second segments of the training data.  I'm aware this probably isn't the most effective approach.\n\nHere's the question - is a score of .74 any better than random guesses + a little luck? \n\nAny thoughts appreciated.\n\n-Rich",
    "2270149": "Hey rich,\nRandom guess would be 0.71, as is the sample submission , reason for that lies in the way this metric is made.\nyour approach isn't the most optimal for sure but it can work, my current LB is following the same approach but with a few tweaks, good luck as there is little time left but you can for sure achieve a better score."
  },
  "source": "meta"
}