{
  "id": 580892,
  "title": "Best single CurveFault_B model?",
  "url": "/competitions/waveform-inversion/discussion/580892",
  "author_name": "",
  "post_date": "2025-05-27T04:28:19.141446500Z",
  "votes": 5,
  "comment_count": 7,
  "views": 0,
  "content": "<h3>CurveFault_B --&gt; Sensor Flip + GaussianNoise --&gt; 25 epochs --&gt; Train: 230 --&gt; CV: 238</h3>\n<hr>\n<blockquote>\n  <p>i'm experimenting only CurveFault_B to test different architectures, what is your best single model trained only on CurveFault_B?</p>\n</blockquote>",
  "messages": [
    {
      "id": "3210348",
      "postDate": "05/27/2025 04:28:19",
      "content": "<h3>CurveFault_B --&gt; Sensor Flip + GaussianNoise --&gt; 25 epochs --&gt; Train: 230 --&gt; CV: 238</h3>\n<hr>\n<blockquote>\n  <p>i'm experimenting only CurveFault_B to test different architectures, what is your best single model trained only on CurveFault_B?</p>\n</blockquote>",
      "rawMarkdown": "### CurveFault_B --> Sensor Flip + GaussianNoise --> 25 epochs --> Train: 230 --> CV: 238\n\n---\n\n> i'm experimenting only CurveFault_B to test different architectures, what is your best single model trained only on CurveFault_B?",
      "votes": null
    },
    {
      "id": "3210499",
      "postDate": "05/27/2025 09:33:04",
      "content": "<p>42.9 on train-set with a generalist model. Val should be a few points higher. E.g. a worser model but with proper train-val split had 58 and 74 respectively.</p>",
      "rawMarkdown": "42.9 on train-set with a generalist model. Val should be a few points higher. E.g. a worser model but with proper train-val split had 58 and 74 respectively.",
      "votes": null
    },
    {
      "id": "3219358",
      "postDate": "06/07/2025 15:11:56",
      "content": "<p><a href=\"https://www.kaggle.com/hoffmanns\" target=\"_blank\">@hoffmanns</a>, is the model trained from backbone weights only on CurveFault_B or from the all dataset pre-trained model </p>\n<ul>\n<li>as my cv not reach below 200 - with different architectures from backbone weights for CurveFault_B.</li>\n</ul>",
      "rawMarkdown": "hoffmanns, is the model trained from backbone weights only on CurveFault_B or from the all dataset pre-trained model \n\n- as my cv not reach below 200 - with different architectures from backbone weights for CurveFault_B.",
      "votes": null
    },
    {
      "id": "3221162",
      "postDate": "06/10/2025 13:46:44",
      "content": "<p>Have you tried the convnext and caformer models in the public notebook.For me, their cv are 206.33 and 259.471, which are trained with 30 epoch. By adjusting some configurations, the cv turn into 140.819 and 150.884.</p>",
      "rawMarkdown": "Have you tried the convnext and caformer models in the public notebook.For me, their cv are 206.33 and 259.471, which are trained with 30 epoch. By adjusting some configurations, the cv turn into 140.819 and 150.884.",
      "votes": null
    },
    {
      "id": "3221167",
      "postDate": "06/10/2025 13:54:00",
      "content": "<p>Yes, i tried Convnext is around ~240, what configurations you mean from 206.33 -&gt; 140.89? <a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> </p>",
      "rawMarkdown": "Yes, i tried Convnext is around ~240, what configurations you mean from 206.33 -> 140.89? @i2nfinit3y",
      "votes": null
    },
    {
      "id": "3221170",
      "postDate": "06/10/2025 13:57:20",
      "content": "<p>I use mix loss function and modify the conv layers in the network.</p>",
      "rawMarkdown": "I use mix loss function and modify the conv layers in the network.",
      "votes": null
    },
    {
      "id": "3221173",
      "postDate": "06/10/2025 14:03:20",
      "content": "<p>Thanks, i will try again <a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> </p>",
      "rawMarkdown": "Thanks, i will try again @i2nfinit3y",
      "votes": null
    },
    {
      "id": "3221434",
      "postDate": "06/11/2025 02:28:28",
      "content": "<p><a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> i tried again thanks, able to reach 60th epoch ~ Train: 58.2,  Val: 79.0</p>\n<table>\n<thead>\n<tr>\n<th>Epoch</th>\n<th>Train (-)</th>\n<th>Val (--)</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>10</td>\n<td>100.3</td>\n<td>94. 7</td>\n</tr>\n<tr>\n<td>20</td>\n<td>84.6</td>\n<td>82.9</td>\n</tr>\n<tr>\n<td>30</td>\n<td>74.0</td>\n<td>81.6</td>\n</tr>\n<tr>\n<td>40</td>\n<td>67.8</td>\n<td>80.0</td>\n</tr>\n<tr>\n<td>60</td>\n<td>58.2</td>\n<td>79.0</td>\n</tr>\n</tbody>\n</table>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F3a035c552ef6becc26868411a229760f%2FScreenshot%202025-06-11%20at%207.59.38AM.png?generation=1749608994487762&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "i2nfinit3y i tried again thanks, able to reach 60th epoch ~ Train: 58.2,  Val: 79.0\n\n| Epoch | Train (-) | Val (--) |\n| --- | --- | --- |\n| 10 | 100.3 | 94. 7 |\n| 20 | 84.6 | 82.9 |\n| 30 | 74.0 | 81.6 |\n| 40 | 67.8 | 80.0 |\n| 60 | 58.2 |  79.0 |\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F3a035c552ef6becc26868411a229760f%2FScreenshot%202025-06-11%20at%207.59.38AM.png?generation=1749608994487762&alt=media)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3210499,
      "author_name": "hoffmanns",
      "author_url": "",
      "post_date": "05/27/2025 09:33:04",
      "content": "<p>42.9 on train-set with a generalist model. Val should be a few points higher. E.g. a worser model but with proper train-val split had 58 and 74 respectively.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3219358,
          "author_name": "seshurajup",
          "author_url": "",
          "post_date": "06/07/2025 15:11:56",
          "content": "<p><a href=\"https://www.kaggle.com/hoffmanns\" target=\"_blank\">@hoffmanns</a>, is the model trained from backbone weights only on CurveFault_B or from the all dataset pre-trained model </p>\n<ul>\n<li>as my cv not reach below 200 - with different architectures from backbone weights for CurveFault_B.</li>\n</ul>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3221162,
      "author_name": "i2nfinit3y",
      "author_url": "",
      "post_date": "06/10/2025 13:46:44",
      "content": "<p>Have you tried the convnext and caformer models in the public notebook.For me, their cv are 206.33 and 259.471, which are trained with 30 epoch. By adjusting some configurations, the cv turn into 140.819 and 150.884.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3221167,
          "author_name": "seshurajup",
          "author_url": "",
          "post_date": "06/10/2025 13:54:00",
          "content": "<p>Yes, i tried Convnext is around ~240, what configurations you mean from 206.33 -&gt; 140.89? <a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> </p>",
          "votes": null,
          "replies": [
            {
              "id": 3221170,
              "author_name": "i2nfinit3y",
              "author_url": "",
              "post_date": "06/10/2025 13:57:20",
              "content": "<p>I use mix loss function and modify the conv layers in the network.</p>",
              "votes": null,
              "replies": [
                {
                  "id": 3221173,
                  "author_name": "seshurajup",
                  "author_url": "",
                  "post_date": "06/10/2025 14:03:20",
                  "content": "<p>Thanks, i will try again <a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> </p>",
                  "votes": null,
                  "replies": []
                },
                {
                  "id": 3221434,
                  "author_name": "seshurajup",
                  "author_url": "",
                  "post_date": "06/11/2025 02:28:28",
                  "content": "<p><a href=\"https://www.kaggle.com/i2nfinit3y\" target=\"_blank\">@i2nfinit3y</a> i tried again thanks, able to reach 60th epoch ~ Train: 58.2,  Val: 79.0</p>\n<table>\n<thead>\n<tr>\n<th>Epoch</th>\n<th>Train (-)</th>\n<th>Val (--)</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>10</td>\n<td>100.3</td>\n<td>94. 7</td>\n</tr>\n<tr>\n<td>20</td>\n<td>84.6</td>\n<td>82.9</td>\n</tr>\n<tr>\n<td>30</td>\n<td>74.0</td>\n<td>81.6</td>\n</tr>\n<tr>\n<td>40</td>\n<td>67.8</td>\n<td>80.0</td>\n</tr>\n<tr>\n<td>60</td>\n<td>58.2</td>\n<td>79.0</td>\n</tr>\n</tbody>\n</table>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F3a035c552ef6becc26868411a229760f%2FScreenshot%202025-06-11%20at%207.59.38AM.png?generation=1749608994487762&amp;alt=media\" alt=\"\"></p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3210348": "### CurveFault_B --> Sensor Flip + GaussianNoise --> 25 epochs --> Train: 230 --> CV: 238\n\n---\n\n> i'm experimenting only CurveFault_B to test different architectures, what is your best single model trained only on CurveFault_B?",
    "3210499": "42.9 on train-set with a generalist model. Val should be a few points higher. E.g. a worser model but with proper train-val split had 58 and 74 respectively.",
    "3219358": "hoffmanns, is the model trained from backbone weights only on CurveFault_B or from the all dataset pre-trained model \n\n- as my cv not reach below 200 - with different architectures from backbone weights for CurveFault_B.",
    "3221162": "Have you tried the convnext and caformer models in the public notebook.For me, their cv are 206.33 and 259.471, which are trained with 30 epoch. By adjusting some configurations, the cv turn into 140.819 and 150.884.",
    "3221167": "Yes, i tried Convnext is around ~240, what configurations you mean from 206.33 -> 140.89? @i2nfinit3y",
    "3221170": "I use mix loss function and modify the conv layers in the network.",
    "3221173": "Thanks, i will try again @i2nfinit3y",
    "3221434": "i2nfinit3y i tried again thanks, able to reach 60th epoch ~ Train: 58.2,  Val: 79.0\n\n| Epoch | Train (-) | Val (--) |\n| --- | --- | --- |\n| 10 | 100.3 | 94. 7 |\n| 20 | 84.6 | 82.9 |\n| 30 | 74.0 | 81.6 |\n| 40 | 67.8 | 80.0 |\n| 60 | 58.2 |  79.0 |\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F3a035c552ef6becc26868411a229760f%2FScreenshot%202025-06-11%20at%207.59.38AM.png?generation=1749608994487762&alt=media)"
  },
  "source": "meta"
}