{
  "id": 410630,
  "title": "How long do your models train for ?",
  "url": "/competitions/birdclef-2023/discussion/410630",
  "author_name": "",
  "post_date": "2023-05-16T02:54:01.303212900Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p><strong>This competition is my first attempt at deep learning, and thus I don't really have any intuition in terms of training time, scores, CV/LB correlation etc.</strong><br>\nSo I made this discussion to be able to build this intuition by learning about other people's experiences rather than my sole own, <br>\nI'm currently getting 0.780 CV 0.77 LB with a training time of 1:30h and 10 minutes per epoch on SED approximately, knowing I am feeding my model already-made spectograms.<br>\nI would appreciate if you could also share some insight <em>(no need to specify the model or anything if you don't want to, whatever information will help me and most likely others as well)</em></p>",
  "messages": [
    {
      "id": "2260965",
      "postDate": "05/16/2023 02:54:01",
      "content": "<p><strong>This competition is my first attempt at deep learning, and thus I don't really have any intuition in terms of training time, scores, CV/LB correlation etc.</strong><br>\nSo I made this discussion to be able to build this intuition by learning about other people's experiences rather than my sole own, <br>\nI'm currently getting 0.780 CV 0.77 LB with a training time of 1:30h and 10 minutes per epoch on SED approximately, knowing I am feeding my model already-made spectograms.<br>\nI would appreciate if you could also share some insight <em>(no need to specify the model or anything if you don't want to, whatever information will help me and most likely others as well)</em></p>",
      "rawMarkdown": "**This competition is my first attempt at deep learning, and thus I don't really have any intuition in terms of training time, scores, CV/LB correlation etc.**\nSo I made this discussion to be able to build this intuition by learning about other people's experiences rather than my sole own, \nI'm currently getting 0.780 CV 0.77 LB with a training time of 1:30h and 10 minutes per epoch on SED approximately, knowing I am feeding my model already-made spectograms.\nI would appreciate if you could also share some insight *(no need to specify the model or anything if you don't want to, whatever information will help me and most likely others as well)*",
      "votes": null
    },
    {
      "id": "2261027",
      "postDate": "05/16/2023 03:59:33",
      "content": "<p>I am mostly in the range of 15 to 20 for the number of epochs and it takes around 30-35 minutes to train one fold. This model scores .8 in the LB.</p>",
      "rawMarkdown": "I am mostly in the range of 15 to 20 for the number of epochs and it takes around 30-35 minutes to train one fold. This model scores .8 in the LB.",
      "votes": null
    },
    {
      "id": "2261393",
      "postDate": "05/16/2023 09:32:09",
      "content": "<p>Did you use previous competition data for pre-training?</p>",
      "rawMarkdown": "Did you use previous competition data for pre-training?",
      "votes": null
    },
    {
      "id": "2261897",
      "postDate": "05/16/2023 15:55:44",
      "content": "<p>Thx for this valuable information ! I made some tweaks and lowered my training time for the same LB score and CV score !</p>",
      "rawMarkdown": "Thx for this valuable information ! I made some tweaks and lowered my training time for the same LB score and CV score !",
      "votes": null
    },
    {
      "id": "2263335",
      "postDate": "05/17/2023 14:07:51",
      "content": "<p>Yes, I am using something similar to public notebooks.</p>",
      "rawMarkdown": "Yes, I am using something similar to public notebooks.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2261027,
      "author_name": "snnclsr",
      "author_url": "",
      "post_date": "05/16/2023 03:59:33",
      "content": "<p>I am mostly in the range of 15 to 20 for the number of epochs and it takes around 30-35 minutes to train one fold. This model scores .8 in the LB.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2261393,
          "author_name": "zifencai",
          "author_url": "",
          "post_date": "05/16/2023 09:32:09",
          "content": "<p>Did you use previous competition data for pre-training?</p>",
          "votes": null,
          "replies": [
            {
              "id": 2263335,
              "author_name": "snnclsr",
              "author_url": "",
              "post_date": "05/17/2023 14:07:51",
              "content": "<p>Yes, I am using something similar to public notebooks.</p>",
              "votes": null,
              "replies": []
            }
          ]
        },
        {
          "id": 2261897,
          "author_name": "janmpia",
          "author_url": "",
          "post_date": "05/16/2023 15:55:44",
          "content": "<p>Thx for this valuable information ! I made some tweaks and lowered my training time for the same LB score and CV score !</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2260965": "**This competition is my first attempt at deep learning, and thus I don't really have any intuition in terms of training time, scores, CV/LB correlation etc.**\nSo I made this discussion to be able to build this intuition by learning about other people's experiences rather than my sole own, \nI'm currently getting 0.780 CV 0.77 LB with a training time of 1:30h and 10 minutes per epoch on SED approximately, knowing I am feeding my model already-made spectograms.\nI would appreciate if you could also share some insight *(no need to specify the model or anything if you don't want to, whatever information will help me and most likely others as well)*",
    "2261027": "I am mostly in the range of 15 to 20 for the number of epochs and it takes around 30-35 minutes to train one fold. This model scores .8 in the LB.",
    "2261393": "Did you use previous competition data for pre-training?",
    "2261897": "Thx for this valuable information ! I made some tweaks and lowered my training time for the same LB score and CV score !",
    "2263335": "Yes, I am using something similar to public notebooks."
  },
  "source": "meta"
}