{
  "id": 209040,
  "title": "This is a missing label.",
  "url": "/competitions/rfcx-species-audio-detection/discussion/209040",
  "author_name": "",
  "post_date": "2021-01-06T03:07:44.880523500Z",
  "votes": 23,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I reported that <a href=\"https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/208830\" target=\"_blank\">discovering missing labels and retraining</a> did not worked.<br>\nFor your inference, I show missing labels example.</p>\n<p>example: <code>0ab6aa734.flac</code></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4704212%2F9ca088bb386abf7114543c019c1d8a5f%2Ffig.png?generation=1609892974092435&amp;alt=media\" alt=\"\"></p>\n<ul>\n<li>Tp label(y_true) is given at around 49~52sec(species id is \"9\").</li>\n<li>But trained model discovered many pseudo-labels(y_predict) in the whole clip.</li>\n</ul>\n<p>Is y_predict true?<br>\n<a href=\"https://www.kaggle.com/shinmurashinmura/this-is-a-missing-label\" target=\"_blank\">Let's listen</a>.</p>\n<p>I repeat, discovering missing labels and retraining <strong>did not improve score</strong> for me.<br>\nGood luck.</p>",
  "messages": [
    {
      "id": "1140472",
      "postDate": "01/06/2021 03:07:44",
      "content": "<p>I reported that <a href=\"https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/208830\" target=\"_blank\">discovering missing labels and retraining</a> did not worked.<br>\nFor your inference, I show missing labels example.</p>\n<p>example: <code>0ab6aa734.flac</code></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4704212%2F9ca088bb386abf7114543c019c1d8a5f%2Ffig.png?generation=1609892974092435&amp;alt=media\" alt=\"\"></p>\n<ul>\n<li>Tp label(y_true) is given at around 49~52sec(species id is \"9\").</li>\n<li>But trained model discovered many pseudo-labels(y_predict) in the whole clip.</li>\n</ul>\n<p>Is y_predict true?<br>\n<a href=\"https://www.kaggle.com/shinmurashinmura/this-is-a-missing-label\" target=\"_blank\">Let's listen</a>.</p>\n<p>I repeat, discovering missing labels and retraining <strong>did not improve score</strong> for me.<br>\nGood luck.</p>",
      "rawMarkdown": "I reported that [discovering missing labels and retraining](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/208830) did not worked.\nFor your inference, I show missing labels example.\n\nexample: ```0ab6aa734.flac```\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4704212%2F9ca088bb386abf7114543c019c1d8a5f%2Ffig.png?generation=1609892974092435&alt=media)\n\n+ Tp label(y_true) is given at around 49~52sec(species id is \"9\").\n+ But trained model discovered many pseudo-labels(y_predict) in the whole clip.\n\nIs y_predict true?\n[Let's listen](https://www.kaggle.com/shinmurashinmura/this-is-a-missing-label).\n\nI repeat, discovering missing labels and retraining **did not improve score** for me.\nGood luck.",
      "votes": null
    },
    {
      "id": "1141349",
      "postDate": "01/06/2021 16:41:10",
      "content": "<p>I think that re-training with weak labels can help, but the weak labels you used, maybe, are too noisy. <br>\nThis is due to the model that overfits the data, but since the validation is very closed to the test, we can not see the gap that usually the overfitting has.</p>",
      "rawMarkdown": "I think that re-training with weak labels can help, but the weak labels you used, maybe, are too noisy. \nThis is due to the model that overfits the data, but since the validation is very closed to the test, we can not see the gap that usually the overfitting has.",
      "votes": null
    },
    {
      "id": "1142889",
      "postDate": "01/07/2021 16:41:06",
      "content": "<p>I do not see this for 0ab6aa734. How do you see 2c6ba715f for species 21 ?</p>",
      "rawMarkdown": "I do not see this for 0ab6aa734. How do you see 2c6ba715f for species 21 ?",
      "votes": null
    },
    {
      "id": "1143723",
      "postDate": "01/08/2021 03:04:08",
      "content": "<p>Thanks, I'll try re-training again with the noisy labels strategy.</p>",
      "rawMarkdown": "Thanks, I'll try re-training again with the noisy labels strategy.",
      "votes": null
    },
    {
      "id": "1144305",
      "postDate": "01/08/2021 11:22:18",
      "content": "<p>Hi, thanks for your great sharing. what is the retraining here exactly? </p>",
      "rawMarkdown": "Hi, thanks for your great sharing. what is the retraining here exactly?",
      "votes": null
    },
    {
      "id": "1144424",
      "postDate": "01/08/2021 13:07:56",
      "content": "<p>I discovered missing labels and retrain with tp and missing labels.</p>",
      "rawMarkdown": "I discovered missing labels and retrain with tp and missing labels.",
      "votes": null
    },
    {
      "id": "1198046",
      "postDate": "02/12/2021 16:53:24",
      "content": "<p>Kind of curious where the visualization of prediction over time came from. Is that from training sample or validation? My predictions dont look nearly as binary, lots more spread over time. </p>",
      "rawMarkdown": "Kind of curious where the visualization of prediction over time came from. Is that from training sample or validation? My predictions dont look nearly as binary, lots more spread over time.",
      "votes": null
    },
    {
      "id": "1198940",
      "postDate": "02/13/2021 12:40:40",
      "content": "<blockquote>\n  <p>Is that from training sample or validation?</p>\n</blockquote>\n<p>It is training sample. Therefore it is binary.<br>\nIf I used validation sample, it may be more spread( not binary).</p>",
      "rawMarkdown": "> Is that from training sample or validation?\n\nIt is training sample. Therefore it is binary.\nIf I used validation sample, it may be more spread( not binary).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1141349,
      "author_name": "guglielmocamporese",
      "author_url": "",
      "post_date": "01/06/2021 16:41:10",
      "content": "<p>I think that re-training with weak labels can help, but the weak labels you used, maybe, are too noisy. <br>\nThis is due to the model that overfits the data, but since the validation is very closed to the test, we can not see the gap that usually the overfitting has.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1143723,
          "author_name": "shinmurashinmura",
          "author_url": "",
          "post_date": "01/08/2021 03:04:08",
          "content": "<p>Thanks, I'll try re-training again with the noisy labels strategy.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1142889,
      "author_name": "glimmung",
      "author_url": "",
      "post_date": "01/07/2021 16:41:06",
      "content": "<p>I do not see this for 0ab6aa734. How do you see 2c6ba715f for species 21 ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1144305,
      "author_name": "mcggood",
      "author_url": "",
      "post_date": "01/08/2021 11:22:18",
      "content": "<p>Hi, thanks for your great sharing. what is the retraining here exactly? </p>",
      "votes": null,
      "replies": [
        {
          "id": 1144424,
          "author_name": "shinmurashinmura",
          "author_url": "",
          "post_date": "01/08/2021 13:07:56",
          "content": "<p>I discovered missing labels and retrain with tp and missing labels.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1198046,
      "author_name": "ryches",
      "author_url": "",
      "post_date": "02/12/2021 16:53:24",
      "content": "<p>Kind of curious where the visualization of prediction over time came from. Is that from training sample or validation? My predictions dont look nearly as binary, lots more spread over time. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1198940,
          "author_name": "shinmurashinmura",
          "author_url": "",
          "post_date": "02/13/2021 12:40:40",
          "content": "<blockquote>\n  <p>Is that from training sample or validation?</p>\n</blockquote>\n<p>It is training sample. Therefore it is binary.<br>\nIf I used validation sample, it may be more spread( not binary).</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1140472": "I reported that [discovering missing labels and retraining](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/208830) did not worked.\nFor your inference, I show missing labels example.\n\nexample: ```0ab6aa734.flac```\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4704212%2F9ca088bb386abf7114543c019c1d8a5f%2Ffig.png?generation=1609892974092435&alt=media)\n\n+ Tp label(y_true) is given at around 49~52sec(species id is \"9\").\n+ But trained model discovered many pseudo-labels(y_predict) in the whole clip.\n\nIs y_predict true?\n[Let's listen](https://www.kaggle.com/shinmurashinmura/this-is-a-missing-label).\n\nI repeat, discovering missing labels and retraining **did not improve score** for me.\nGood luck.",
    "1141349": "I think that re-training with weak labels can help, but the weak labels you used, maybe, are too noisy. \nThis is due to the model that overfits the data, but since the validation is very closed to the test, we can not see the gap that usually the overfitting has.",
    "1142889": "I do not see this for 0ab6aa734. How do you see 2c6ba715f for species 21 ?",
    "1143723": "Thanks, I'll try re-training again with the noisy labels strategy.",
    "1144305": "Hi, thanks for your great sharing. what is the retraining here exactly?",
    "1144424": "I discovered missing labels and retrain with tp and missing labels.",
    "1198046": "Kind of curious where the visualization of prediction over time came from. Is that from training sample or validation? My predictions dont look nearly as binary, lots more spread over time.",
    "1198940": "> Is that from training sample or validation?\n\nIt is training sample. Therefore it is binary.\nIf I used validation sample, it may be more spread( not binary)."
  },
  "source": "meta"
}