{
  "id": 548832,
  "title": "Only autoencoder model has LB0.38",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/548832",
  "author_name": "",
  "post_date": "2024-11-29T05:27:19.410951700Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>This is my first post.</p>\n<p>For the past few days I have been working my heart out on extracting features such as autoencoders and 1D CNN VAE from accelerometer data.</p>\n<p>However, as you can see from trying the top notebook, the LB drops significantly to 0.38 when trying only the model with autoencoders.</p>\n<p>Many of the copied and pasted notebooks are ensemble with autoencoder use + regular boosting tree x 2, but this score seems to be an add-on measure that blends the extreme predictions generated by the autoencoder with the regular boosting tree model.</p>\n<p>If we use a neural network, wouldn't we have to specify the detection pattern in detail?</p>",
  "messages": [
    {
      "id": "3058141",
      "postDate": "11/29/2024 05:27:19",
      "content": "<p>This is my first post.</p>\n<p>For the past few days I have been working my heart out on extracting features such as autoencoders and 1D CNN VAE from accelerometer data.</p>\n<p>However, as you can see from trying the top notebook, the LB drops significantly to 0.38 when trying only the model with autoencoders.</p>\n<p>Many of the copied and pasted notebooks are ensemble with autoencoder use + regular boosting tree x 2, but this score seems to be an add-on measure that blends the extreme predictions generated by the autoencoder with the regular boosting tree model.</p>\n<p>If we use a neural network, wouldn't we have to specify the detection pattern in detail?</p>",
      "rawMarkdown": "This is my first post.\n\nFor the past few days I have been working my heart out on extracting features such as autoencoders and 1D CNN VAE from accelerometer data.\n\nHowever, as you can see from trying the top notebook, the LB drops significantly to 0.38 when trying only the model with autoencoders.\n\nMany of the copied and pasted notebooks are ensemble with autoencoder use + regular boosting tree x 2, but this score seems to be an add-on measure that blends the extreme predictions generated by the autoencoder with the regular boosting tree model.\n\n\nIf we use a neural network, wouldn't we have to specify the detection pattern in detail?",
      "votes": null
    },
    {
      "id": "3061025",
      "postDate": "12/02/2024 10:40:21",
      "content": "<p>Good observation for your first post <a href=\"https://www.kaggle.com/sfukushi\" target=\"_blank\">@sfukushi</a> <br>\nThis itself tells us how god the autoencoder is for this challenge!</p>",
      "rawMarkdown": "Good observation for your first post @sfukushi \nThis itself tells us how god the autoencoder is for this challenge!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3061025,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "12/02/2024 10:40:21",
      "content": "<p>Good observation for your first post <a href=\"https://www.kaggle.com/sfukushi\" target=\"_blank\">@sfukushi</a> <br>\nThis itself tells us how god the autoencoder is for this challenge!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3058141": "This is my first post.\n\nFor the past few days I have been working my heart out on extracting features such as autoencoders and 1D CNN VAE from accelerometer data.\n\nHowever, as you can see from trying the top notebook, the LB drops significantly to 0.38 when trying only the model with autoencoders.\n\nMany of the copied and pasted notebooks are ensemble with autoencoder use + regular boosting tree x 2, but this score seems to be an add-on measure that blends the extreme predictions generated by the autoencoder with the regular boosting tree model.\n\n\nIf we use a neural network, wouldn't we have to specify the detection pattern in detail?",
    "3061025": "Good observation for your first post @sfukushi \nThis itself tells us how god the autoencoder is for this challenge!"
  },
  "source": "meta"
}