{
  "id": 177584,
  "title": "Test set prob results ",
  "url": "/competitions/birdsong-recognition/discussion/177584",
  "author_name": "",
  "post_date": "2020-08-26T13:43:27.636012800Z",
  "votes": 15,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Public and Private test data combined contains exactly 150 audio files. <br>\n50 audio files under each site 1,2&amp;3. <br>\nSince site 1 and 2 need predictions of every  5 seconds and site 3 clip level prediction.<br>\nconsidering each audio file is of 10 minutes or 600 seconds , each file will need 600/5=120 predictions. <br>\nTotal predictions required:<br>\nsite 1 &amp; 2 = 120*100 = 12,000<br>\nsite 3=50 </p>\n<p>So skipping site 3 files may give more time to for running multiple models on site 1 &amp; 2 files for better ensembles results.</p>\n<p>Count unique audio filenames, put some conditional statements in the submission notebook to verify. <br>\nUpdate : Made a notebook to demonstrate the above probing <a href=\"https://www.kaggle.com/rakshith1/probing-test-set?scriptVersionId=41474057\" target=\"_blank\">here</a>.                  </p>",
  "messages": [
    {
      "id": "986431",
      "postDate": "08/26/2020 13:43:27",
      "content": "<p>Public and Private test data combined contains exactly 150 audio files. <br>\n50 audio files under each site 1,2&amp;3. <br>\nSince site 1 and 2 need predictions of every  5 seconds and site 3 clip level prediction.<br>\nconsidering each audio file is of 10 minutes or 600 seconds , each file will need 600/5=120 predictions. <br>\nTotal predictions required:<br>\nsite 1 &amp; 2 = 120*100 = 12,000<br>\nsite 3=50 </p>\n<p>So skipping site 3 files may give more time to for running multiple models on site 1 &amp; 2 files for better ensembles results.</p>\n<p>Count unique audio filenames, put some conditional statements in the submission notebook to verify. <br>\nUpdate : Made a notebook to demonstrate the above probing <a href=\"https://www.kaggle.com/rakshith1/probing-test-set?scriptVersionId=41474057\" target=\"_blank\">here</a>.                  </p>",
      "rawMarkdown": "Public and Private test data combined contains exactly 150 audio files. \n50 audio files under each site 1,2&3. \nSince site 1 and 2 need predictions of every  5 seconds and site 3 clip level prediction.\nconsidering each audio file is of 10 minutes or 600 seconds , each file will need 600/5=120 predictions. \nTotal predictions required:\nsite 1 & 2 = 120*100 = 12,000\nsite 3=50 \n\nSo skipping site 3 files may give more time to for running multiple models on site 1 & 2 files for better ensembles results.\n \nCount unique audio filenames, put some conditional statements in the submission notebook to verify. \nUpdate : Made a notebook to demonstrate the above probing [here](https://www.kaggle.com/rakshith1/probing-test-set?scriptVersionId=41474057).",
      "votes": null
    },
    {
      "id": "986472",
      "postDate": "08/26/2020 14:26:19",
      "content": "<p>50 audio files under each site 1,2&amp;3.  --- how do you know this?</p>",
      "rawMarkdown": "50 audio files under each site 1,2&3.  --- how do you know this?",
      "votes": null
    },
    {
      "id": "987388",
      "postDate": "08/27/2020 08:09:08",
      "content": "<p>Attached a notebook in the post showing how it is done. <br>\nCheers. </p>",
      "rawMarkdown": "Attached a notebook in the post showing how it is done. \nCheers.",
      "votes": null
    },
    {
      "id": "989017",
      "postDate": "08/28/2020 13:53:07",
      "content": "<p>Thank you. Great to know. I was thinking to do something like this but was lazy and did other experiments.</p>",
      "rawMarkdown": "Thank you. Great to know. I was thinking to do something like this but was lazy and did other experiments.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 986472,
      "author_name": "yaroshevskiy",
      "author_url": "",
      "post_date": "08/26/2020 14:26:19",
      "content": "<p>50 audio files under each site 1,2&amp;3.  --- how do you know this?</p>",
      "votes": null,
      "replies": [
        {
          "id": 987388,
          "author_name": "rakshith1",
          "author_url": "",
          "post_date": "08/27/2020 08:09:08",
          "content": "<p>Attached a notebook in the post showing how it is done. <br>\nCheers. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 989017,
      "author_name": "leodav",
      "author_url": "",
      "post_date": "08/28/2020 13:53:07",
      "content": "<p>Thank you. Great to know. I was thinking to do something like this but was lazy and did other experiments.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "986431": "Public and Private test data combined contains exactly 150 audio files. \n50 audio files under each site 1,2&3. \nSince site 1 and 2 need predictions of every  5 seconds and site 3 clip level prediction.\nconsidering each audio file is of 10 minutes or 600 seconds , each file will need 600/5=120 predictions. \nTotal predictions required:\nsite 1 & 2 = 120*100 = 12,000\nsite 3=50 \n\nSo skipping site 3 files may give more time to for running multiple models on site 1 & 2 files for better ensembles results.\n \nCount unique audio filenames, put some conditional statements in the submission notebook to verify. \nUpdate : Made a notebook to demonstrate the above probing [here](https://www.kaggle.com/rakshith1/probing-test-set?scriptVersionId=41474057).",
    "986472": "50 audio files under each site 1,2&3.  --- how do you know this?",
    "987388": "Attached a notebook in the post showing how it is done. \nCheers.",
    "989017": "Thank you. Great to know. I was thinking to do something like this but was lazy and did other experiments."
  },
  "source": "meta"
}