{
  "id": 239173,
  "title": "6 Vs 3 channels",
  "url": "/competitions/seti-breakthrough-listen/discussion/239173",
  "author_name": "Awsaf",
  "post_date": "2021-05-15T05:37:47.170000",
  "votes": 21,
  "comment_count": 2,
  "views": 0,
  "content": "<h2>Update - 18/05/2021</h2>\n<p>I've mistakenly forgotten to update the notebook for <strong>3</strong> channels. After the update, it seems that <strong>3</strong> channel is a bit better if we consider <strong>CV</strong>.</p>\n<p><img src=\"https://i.ibb.co/F4dvB7z/6-vs-3.png\" alt=\"6-vs-3\"><br>\nThis notebook is an extension of this <a href=\"https://www.kaggle.com/c/seti-breakthrough-listen/discussion/238298\" target=\"_blank\">discussion</a>. This experiment is to check if only <strong>3</strong> channels can pull off <strong>6</strong> channels score.</p>\n<p>For this experiment, I used the following two notebooks,</p>\n<ul>\n<li><strong>6</strong> channels: <a href=\"https://www.kaggle.com/yasufuminakama/seti-nfnet-l0-starter-training\" target=\"_blank\">SETI / NFNet_l0 starter [training]</a> by <a href=\"https://www.kaggle.com/yasufuminakama\" target=\"_blank\">@yasufuminakama</a>.</li>\n<li><strong>3</strong> channels: <a href=\"https://www.kaggle.com/awsaf49/seti-bl-nfnet-l0-3-channels\" target=\"_blank\">SETI-BL: NFNet_l0 3 Channels</a></li>\n</ul>\n<blockquote>\n  <p>So, can only <strong>3</strong> channels can pull off <strong>6</strong> channels score [<strong>CV:0.968 | LB:0.95</strong>]?</p>\n</blockquote>\n<h3>Result:</h3>\n<table>\n<thead>\n<tr>\n<th>channels</th>\n<th>CV</th>\n<th>LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>6</td>\n<td>0.968</td>\n<td>0.95</td>\n</tr>\n<tr>\n<td>3</td>\n<td>0.970</td>\n<td>0.96</td>\n</tr>\n</tbody>\n</table>",
  "messages": [
    {
      "id": 1308264,
      "postDate": "2021-05-15T05:37:47.170Z",
      "content": "<h2>Update - 18/05/2021</h2>\n<p>I've mistakenly forgotten to update the notebook for <strong>3</strong> channels. After the update, it seems that <strong>3</strong> channel is a bit better if we consider <strong>CV</strong>.</p>\n<p><img src=\"https://i.ibb.co/F4dvB7z/6-vs-3.png\" alt=\"6-vs-3\"><br>\nThis notebook is an extension of this <a href=\"https://www.kaggle.com/c/seti-breakthrough-listen/discussion/238298\" target=\"_blank\">discussion</a>. This experiment is to check if only <strong>3</strong> channels can pull off <strong>6</strong> channels score.</p>\n<p>For this experiment, I used the following two notebooks,</p>\n<ul>\n<li><strong>6</strong> channels: <a href=\"https://www.kaggle.com/yasufuminakama/seti-nfnet-l0-starter-training\" target=\"_blank\">SETI / NFNet_l0 starter [training]</a> by <a href=\"https://www.kaggle.com/yasufuminakama\" target=\"_blank\">@yasufuminakama</a>.</li>\n<li><strong>3</strong> channels: <a href=\"https://www.kaggle.com/awsaf49/seti-bl-nfnet-l0-3-channels\" target=\"_blank\">SETI-BL: NFNet_l0 3 Channels</a></li>\n</ul>\n<blockquote>\n  <p>So, can only <strong>3</strong> channels can pull off <strong>6</strong> channels score [<strong>CV:0.968 | LB:0.95</strong>]?</p>\n</blockquote>\n<h3>Result:</h3>\n<table>\n<thead>\n<tr>\n<th>channels</th>\n<th>CV</th>\n<th>LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>6</td>\n<td>0.968</td>\n<td>0.95</td>\n</tr>\n<tr>\n<td>3</td>\n<td>0.970</td>\n<td>0.96</td>\n</tr>\n</tbody>\n</table>",
      "rawMarkdown": "## Update - 18/05/2021\nI've mistakenly forgotten to update the notebook for **3** channels. After the update, it seems that **3** channel is a bit better if we consider **CV**.\n\n<img src=\"https://i.ibb.co/F4dvB7z/6-vs-3.png\" alt=\"6-vs-3\" border=\"0\" width=\"600\" height=\"400\">\nThis notebook is an extension of this [discussion](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/238298). This experiment is to check if only **3** channels can pull off **6** channels score.\n\nFor this experiment, I used the following two notebooks,\n* **6** channels: [SETI / NFNet_l0 starter [training]](https://www.kaggle.com/yasufuminakama/seti-nfnet-l0-starter-training) by @yasufuminakama.\n* **3** channels: [SETI-BL: NFNet_l0 3 Channels](https://www.kaggle.com/awsaf49/seti-bl-nfnet-l0-3-channels)\n> So, can only **3** channels can pull off **6** channels score [**CV:0.968 | LB:0.95**]?\n\n### Result:\n\n| channels | CV | LB |\n| --- | --- | --- |\n| 6 | 0.968 | 0.95 |\n| 3 | 0.970 | 0.96 |\n",
      "votes": 21
    },
    {
      "id": 1310756,
      "postDate": "2021-05-16T23:21:16.707Z",
      "content": "<p>I always trust a Naruto fan. High five for the sage of six paths.</p>\n<p>Anyway nice experiment. But in my opinion, it is not shocking. We know for <strong>sure</strong> that even number spectrograms don't have any UFO signal-related information. </p>\n<p>The most obvious formulation of the problem in my opinion is anomaly detection. Since the even-numbered spectrograms are generated when the antenna is off-target, it's semantically different. The only obvious information that it carries is the presence of interference and from your experiment I would conclude that this information wasn't that useful. I might be wrong. </p>",
      "rawMarkdown": "I always trust a Naruto fan. High five for the sage of six paths.\n\nAnyway nice experiment. But in my opinion, it is not shocking. We know for **sure** that even number spectrograms don't have any UFO signal-related information. \n\nThe most obvious formulation of the problem in my opinion is anomaly detection. Since the even-numbered spectrograms are generated when the antenna is off-target, it's semantically different. The only obvious information that it carries is the presence of interference and from your experiment I would conclude that this information wasn't that useful. I might be wrong. ",
      "votes": 1
    },
    {
      "id": 1308641,
      "postDate": "2021-05-15T10:54:39.103Z",
      "content": "<p>I tried 3 channels in a different ways and also my initial results are almost similar to 6 channels.</p>",
      "rawMarkdown": "I tried 3 channels in a different ways and also my initial results are almost similar to 6 channels.",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1310756,
      "author_name": "Ayush Thakur",
      "author_url": "",
      "post_date": "2021-05-16T23:21:16.707000",
      "content": "<p>I always trust a Naruto fan. High five for the sage of six paths.</p>\n<p>Anyway nice experiment. But in my opinion, it is not shocking. We know for <strong>sure</strong> that even number spectrograms don't have any UFO signal-related information. </p>\n<p>The most obvious formulation of the problem in my opinion is anomaly detection. Since the even-numbered spectrograms are generated when the antenna is off-target, it's semantically different. The only obvious information that it carries is the presence of interference and from your experiment I would conclude that this information wasn't that useful. I might be wrong. </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1308641,
      "author_name": "Aman Deep Gupta",
      "author_url": "",
      "post_date": "2021-05-15T10:54:39.103000",
      "content": "<p>I tried 3 channels in a different ways and also my initial results are almost similar to 6 channels.</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1308264": "## Update - 18/05/2021\nI've mistakenly forgotten to update the notebook for **3** channels. After the update, it seems that **3** channel is a bit better if we consider **CV**.\n\n<img src=\"https://i.ibb.co/F4dvB7z/6-vs-3.png\" alt=\"6-vs-3\" border=\"0\" width=\"600\" height=\"400\">\nThis notebook is an extension of this [discussion](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/238298). This experiment is to check if only **3** channels can pull off **6** channels score.\n\nFor this experiment, I used the following two notebooks,\n* **6** channels: [SETI / NFNet_l0 starter [training]](https://www.kaggle.com/yasufuminakama/seti-nfnet-l0-starter-training) by @yasufuminakama.\n* **3** channels: [SETI-BL: NFNet_l0 3 Channels](https://www.kaggle.com/awsaf49/seti-bl-nfnet-l0-3-channels)\n> So, can only **3** channels can pull off **6** channels score [**CV:0.968 | LB:0.95**]?\n\n### Result:\n\n| channels | CV | LB |\n| --- | --- | --- |\n| 6 | 0.968 | 0.95 |\n| 3 | 0.970 | 0.96 |\n",
    "1310756": "I always trust a Naruto fan. High five for the sage of six paths.\n\nAnyway nice experiment. But in my opinion, it is not shocking. We know for **sure** that even number spectrograms don't have any UFO signal-related information. \n\nThe most obvious formulation of the problem in my opinion is anomaly detection. Since the even-numbered spectrograms are generated when the antenna is off-target, it's semantically different. The only obvious information that it carries is the presence of interference and from your experiment I would conclude that this information wasn't that useful. I might be wrong. ",
    "1308641": "I tried 3 channels in a different ways and also my initial results are almost similar to 6 channels."
  }
}