{
  "id": 275405,
  "title": "11th place brief solution",
  "url": "/competitions/g2net-gravitational-wave-detection/writeups/hoyeol-sohn-11th-place-brief-solution",
  "author_name": "",
  "post_date": "2021-09-30T09:34:24.090Z",
  "votes": 17,
  "comment_count": 1,
  "views": 0,
  "content": "<p>hi, im a student learning Computer science/Statistics at university and im new to Kaggle.✋<br>\nI don't really know much about GW/signal processing/ or other deep learning skills(augmentation, stacking,etc) so I just mainly stuck to modeling(1D CNNs).</p>\n<p><strong>preprocessing:</strong><br>\n1) bpf 30-500<br>\n2) bpf 25-1020</p>\n<p><strong>augmentation:</strong><br>\nmixup</p>\n<p><strong>models:</strong><br>\nsingle 1DCNN model v1 -&gt; cv 8800 lb 8800<br>\nsingle 1DCNN model v2 -&gt; cv 877x lb ?(didn't try)</p>\n<p>v1 + v2 average ensemble -&gt; cv 882x public lb 882x (!)<br>\nv1 + v2 with preprocessing1,2 -&gt; cv 883x public lb 8830<br>\nusing whole dataset -&gt; public lb8833<br>\nstacking with a single cqt model in public notebook(cv875x, thanks to <a href=\"https://www.kaggle.com/ragnar123\" target=\"_blank\">@ragnar123</a>)<br>\n-&gt; public lb8836</p>\n<p>for 1DCNN models, I used residual connection and dilated convolution/standard convolution with ~24 layers, ~2M parameters.<br>\nI tried to figure out why v1+v2 ensembling gets so much improvement and tried to build a model with both models' advantages but I couldn't due to lack of time.</p>\n<p>so glad for learning so many things/achieving a gold medal in my very first DL competition!<br>\nAnd special thanks to <a href=\"https://www.kaggle.com/hidehisaarai1213\" target=\"_blank\">@hidehisaarai1213</a> and <a href=\"https://www.kaggle.com/miklgr500\" target=\"_blank\">@miklgr500</a> for providing such a great notebook to start with:)</p>",
  "messages": [
    {
      "id": "1529277",
      "postDate": "09/30/2021 09:06:35",
      "content": "<p>hi, im a student learning Computer science/Statistics at university and im new to Kaggle.✋<br>\nI don't really know much about GW/signal processing/ or other deep learning skills(augmentation, stacking,etc) so I just mainly stuck to modeling(1D CNNs).</p>\n<p><strong>preprocessing:</strong><br>\n1) bpf 30-500<br>\n2) bpf 25-1020</p>\n<p><strong>augmentation:</strong><br>\nmixup</p>\n<p><strong>models:</strong><br>\nsingle 1DCNN model v1 -&gt; cv 8800 lb 8800<br>\nsingle 1DCNN model v2 -&gt; cv 877x lb ?(didn't try)</p>\n<p>v1 + v2 average ensemble -&gt; cv 882x public lb 882x (!)<br>\nv1 + v2 with preprocessing1,2 -&gt; cv 883x public lb 8830<br>\nusing whole dataset -&gt; public lb8833<br>\nstacking with a single cqt model in public notebook(cv875x, thanks to <a href=\"https://www.kaggle.com/ragnar123\" target=\"_blank\">@ragnar123</a>)<br>\n-&gt; public lb8836</p>\n<p>for 1DCNN models, I used residual connection and dilated convolution/standard convolution with ~24 layers, ~2M parameters.<br>\nI tried to figure out why v1+v2 ensembling gets so much improvement and tried to build a model with both models' advantages but I couldn't due to lack of time.</p>\n<p>so glad for learning so many things/achieving a gold medal in my very first DL competition!<br>\nAnd special thanks to <a href=\"https://www.kaggle.com/hidehisaarai1213\" target=\"_blank\">@hidehisaarai1213</a> and <a href=\"https://www.kaggle.com/miklgr500\" target=\"_blank\">@miklgr500</a> for providing such a great notebook to start with:)</p>",
      "rawMarkdown": "hi, im a student learning Computer science/Statistics at university and im new to Kaggle.✋\nI don't really know much about GW/signal processing/ or other deep learning skills(augmentation, stacking,etc) so I just mainly stuck to modeling(1D CNNs).\n\n**preprocessing:**\n1) bpf 30-500\n2) bpf 25-1020\n\n**augmentation:**\nmixup\n\n**models:**\nsingle 1DCNN model v1 -> cv 8800 lb 8800\nsingle 1DCNN model v2 -> cv 877x lb ?(didn't try)\n\nv1 + v2 average ensemble -> cv 882x public lb 882x (!)\nv1 + v2 with preprocessing1,2 -> cv 883x public lb 8830\nusing whole dataset -> public lb8833\nstacking with a single cqt model in public notebook(cv875x, thanks to @ragnar123)\n-> public lb8836\n\nfor 1DCNN models, I used residual connection and dilated convolution/standard convolution with ~24 layers, ~2M parameters.\nI tried to figure out why v1+v2 ensembling gets so much improvement and tried to build a model with both models' advantages but I couldn't due to lack of time.\n\nso glad for learning so many things/achieving a gold medal in my very first DL competition!\nAnd special thanks to @hidehisaarai1213 and @miklgr500 for providing such a great notebook to start with:)",
      "votes": null
    },
    {
      "id": "1559881",
      "postDate": "10/27/2021 08:05:29",
      "content": "<p>Hey All,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "rawMarkdown": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1559881,
      "author_name": "zerafachris",
      "author_url": "",
      "post_date": "10/27/2021 08:05:29",
      "content": "<p>Hey All,</p>\n<p>Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey <a href=\"https://forms.gle/QP9L16niPexozyhu5\" target=\"_blank\">https://forms.gle/QP9L16niPexozyhu5</a>.</p>\n<p>Thank you all,</p>\n<p>Regards,<br>\nChris</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1529277": "hi, im a student learning Computer science/Statistics at university and im new to Kaggle.✋\nI don't really know much about GW/signal processing/ or other deep learning skills(augmentation, stacking,etc) so I just mainly stuck to modeling(1D CNNs).\n\n**preprocessing:**\n1) bpf 30-500\n2) bpf 25-1020\n\n**augmentation:**\nmixup\n\n**models:**\nsingle 1DCNN model v1 -> cv 8800 lb 8800\nsingle 1DCNN model v2 -> cv 877x lb ?(didn't try)\n\nv1 + v2 average ensemble -> cv 882x public lb 882x (!)\nv1 + v2 with preprocessing1,2 -> cv 883x public lb 8830\nusing whole dataset -> public lb8833\nstacking with a single cqt model in public notebook(cv875x, thanks to @ragnar123)\n-> public lb8836\n\nfor 1DCNN models, I used residual connection and dilated convolution/standard convolution with ~24 layers, ~2M parameters.\nI tried to figure out why v1+v2 ensembling gets so much improvement and tried to build a model with both models' advantages but I couldn't due to lack of time.\n\nso glad for learning so many things/achieving a gold medal in my very first DL competition!\nAnd special thanks to @hidehisaarai1213 and @miklgr500 for providing such a great notebook to start with:)",
    "1559881": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris"
  },
  "source": "meta"
}