{
  "id": 459757,
  "title": "Solution - Place 154",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/459757",
  "author_name": "Sietse Schröder",
  "post_date": "2023-12-06T14:11:56.467000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Our solution was the mean ensemble over the Gaussian outputs of the CNN Spectogram approach posted by <a href=\"https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states\" target=\"_blank\">tubo213</a> . This is the <a href=\"https://www.kaggle.com/code/sietseschrder/cmi-ensemble-inference-lb-0-736\" target=\"_blank\">link</a> to our submission notebook. </p>\n<p>The ensembled models:</p>\n<ul>\n<li>Resnet 34, batch size 16, 55 epochs</li>\n<li>Resnet 34, batch size 8, 55 epochs</li>\n<li>Resnet 34, batch size 4, 55 epochs</li>\n<li>Resnet 50, batch size 16, 50 epochs</li>\n</ul>\n<p>One of the obvious limitations of this approach is the lack of diversity in the ensembled models. The best individual model scored 0.726.</p>",
  "messages": [
    {
      "id": 2551123,
      "postDate": "2023-12-06T14:11:56.467Z",
      "content": "<p>Our solution was the mean ensemble over the Gaussian outputs of the CNN Spectogram approach posted by <a href=\"https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states\" target=\"_blank\">tubo213</a> . This is the <a href=\"https://www.kaggle.com/code/sietseschrder/cmi-ensemble-inference-lb-0-736\" target=\"_blank\">link</a> to our submission notebook. </p>\n<p>The ensembled models:</p>\n<ul>\n<li>Resnet 34, batch size 16, 55 epochs</li>\n<li>Resnet 34, batch size 8, 55 epochs</li>\n<li>Resnet 34, batch size 4, 55 epochs</li>\n<li>Resnet 50, batch size 16, 50 epochs</li>\n</ul>\n<p>One of the obvious limitations of this approach is the lack of diversity in the ensembled models. The best individual model scored 0.726.</p>",
      "rawMarkdown": "Our solution was the mean ensemble over the Gaussian outputs of the CNN Spectogram approach posted by [tubo213](https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states) . This is the [link](https://www.kaggle.com/code/sietseschrder/cmi-ensemble-inference-lb-0-736) to our submission notebook. \n\n\nThe ensembled models:\n- Resnet 34, batch size 16, 55 epochs\n- Resnet 34, batch size 8, 55 epochs\n- Resnet 34, batch size 4, 55 epochs\n- Resnet 50, batch size 16, 50 epochs\n\nOne of the obvious limitations of this approach is the lack of diversity in the ensembled models. The best individual model scored 0.726.",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2551123": "Our solution was the mean ensemble over the Gaussian outputs of the CNN Spectogram approach posted by [tubo213](https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states) . This is the [link](https://www.kaggle.com/code/sietseschrder/cmi-ensemble-inference-lb-0-736) to our submission notebook. \n\n\nThe ensembled models:\n- Resnet 34, batch size 16, 55 epochs\n- Resnet 34, batch size 8, 55 epochs\n- Resnet 34, batch size 4, 55 epochs\n- Resnet 50, batch size 16, 50 epochs\n\nOne of the obvious limitations of this approach is the lack of diversity in the ensembled models. The best individual model scored 0.726."
  }
}