{
  "id": 238522,
  "title": "Best Model Experiment",
  "url": "/competitions/seti-breakthrough-listen/discussion/238522",
  "author_name": "Salman",
  "post_date": "2021-05-12T12:48:21.315000",
  "votes": 8,
  "comment_count": 4,
  "views": 0,
  "content": "<p>-EffiecientNet-B01<br>\n-No Resize<br>\n-No Augmentation<br>\n-BCE Loss<br>\n-5 Folds<br>\n-CV : 0.96 LB : 0.96</p>\n<p><a href=\"https://www.kaggle.com/micheomaano/simple-yet-effectiveb0\" target=\"_blank\">Notebook</a></p>\n<p>My best model till now is based on TTA + Folds + Sampling + Resize</p>\n<p><strong>Things that didn't work for me till now (But I'll try to do more experiments)</strong></p>\n<p>ConvLSTM<br>\nUnderSampling<br>\nFocal Loss<br>\nPseudo Labelling</p>\n<p>I guess, if we can share ideas to solve this problem then it would be a great contribution. <br>\nRight now I am trying to find a way to normalise frequencies for convolutions. I believe a good representation of frequencies is important in this case.</p>",
  "messages": [
    {
      "id": 1304129,
      "postDate": "2021-05-12T12:48:21.317Z",
      "content": "<p>-EffiecientNet-B01<br>\n-No Resize<br>\n-No Augmentation<br>\n-BCE Loss<br>\n-5 Folds<br>\n-CV : 0.96 LB : 0.96</p>\n<p><a href=\"https://www.kaggle.com/micheomaano/simple-yet-effectiveb0\" target=\"_blank\">Notebook</a></p>\n<p>My best model till now is based on TTA + Folds + Sampling + Resize</p>\n<p><strong>Things that didn't work for me till now (But I'll try to do more experiments)</strong></p>\n<p>ConvLSTM<br>\nUnderSampling<br>\nFocal Loss<br>\nPseudo Labelling</p>\n<p>I guess, if we can share ideas to solve this problem then it would be a great contribution. <br>\nRight now I am trying to find a way to normalise frequencies for convolutions. I believe a good representation of frequencies is important in this case.</p>",
      "rawMarkdown": "-EffiecientNet-B01\n-No Resize\n-No Augmentation\n-BCE Loss\n-5 Folds\n-CV : 0.96 LB : 0.96\n\n[Notebook](https://www.kaggle.com/micheomaano/simple-yet-effectiveb0 )\n\nMy best model till now is based on TTA + Folds + Sampling + Resize\n\n**Things that didn't work for me till now (But I'll try to do more experiments)**\n\nConvLSTM\nUnderSampling\nFocal Loss\nPseudo Labelling\n\nI guess, if we can share ideas to solve this problem then it would be a great contribution. \nRight now I am trying to find a way to normalise frequencies for convolutions. I believe a good representation of frequencies is important in this case.",
      "votes": 8
    },
    {
      "id": 1304133,
      "postDate": "2021-05-12T12:52:10.033Z",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/micheomaano\" target=\"_blank\">@micheomaano</a>. The hyperlink redirects to the wrong page. Could you recheck please?</p>",
      "rawMarkdown": "Hi @micheomaano. The hyperlink redirects to the wrong page. Could you recheck please?",
      "replies": [
        {
          "id": 1304136,
          "postDate": "2021-05-12T12:52:59.953Z",
          "content": "<p>Updated 👍</p>",
          "rawMarkdown": "Updated 👍",
          "votes": 1
        },
        {
          "id": 1304146,
          "postDate": "2021-05-12T12:56:49.867Z",
          "content": "<p>Simple yet elegant👌. But it seems that it's a quick saved version. Did you use another notebook for submission?</p>",
          "rawMarkdown": "Simple yet elegant👌. But it seems that it's a quick saved version. Did you use another notebook for submission?"
        },
        {
          "id": 1304175,
          "postDate": "2021-05-12T13:20:15.577Z",
          "content": "<p>No. I trained on my workstation. </p>",
          "rawMarkdown": "No. I trained on my workstation. ",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1304133,
      "author_name": "Awsaf",
      "author_url": "",
      "post_date": "2021-05-12T12:52:10.033000",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/micheomaano\" target=\"_blank\">@micheomaano</a>. The hyperlink redirects to the wrong page. Could you recheck please?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1304136,
          "author_name": "Salman",
          "author_url": "",
          "post_date": "2021-05-12T12:52:59.953000",
          "content": "<p>Updated 👍</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1304146,
          "author_name": "Awsaf",
          "author_url": "",
          "post_date": "2021-05-12T12:56:49.867000",
          "content": "<p>Simple yet elegant👌. But it seems that it's a quick saved version. Did you use another notebook for submission?</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1304175,
          "author_name": "Salman",
          "author_url": "",
          "post_date": "2021-05-12T13:20:15.577000",
          "content": "<p>No. I trained on my workstation. </p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1304129": "-EffiecientNet-B01\n-No Resize\n-No Augmentation\n-BCE Loss\n-5 Folds\n-CV : 0.96 LB : 0.96\n\n[Notebook](https://www.kaggle.com/micheomaano/simple-yet-effectiveb0 )\n\nMy best model till now is based on TTA + Folds + Sampling + Resize\n\n**Things that didn't work for me till now (But I'll try to do more experiments)**\n\nConvLSTM\nUnderSampling\nFocal Loss\nPseudo Labelling\n\nI guess, if we can share ideas to solve this problem then it would be a great contribution. \nRight now I am trying to find a way to normalise frequencies for convolutions. I believe a good representation of frequencies is important in this case.",
    "1304133": "Hi @micheomaano. The hyperlink redirects to the wrong page. Could you recheck please?"
  }
}