{
  "id": 541677,
  "title": "NNS models VS lgb, cat, xgb models",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/541677",
  "author_name": "",
  "post_date": "2024-10-20T20:05:43.550500400Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>when we can prefer NNs models vs modern models for small size of tabular data?</p>",
  "messages": [
    {
      "id": "3023663",
      "postDate": "10/20/2024 20:05:43",
      "content": "<p>when we can prefer NNs models vs modern models for small size of tabular data?</p>",
      "rawMarkdown": "when we can prefer NNs models vs modern models for small size of tabular data?",
      "votes": null
    },
    {
      "id": "3027322",
      "postDate": "10/24/2024 17:14:08",
      "content": "<p>NN usually won't perform as well as the tree-based boosting methods like XGB, CatBoost and LGBM with small-sized tabular data (like less than 400k rows). It doesn't mean they can be competitive, you just have to experiment with a variety of architectures and hyperparameters.</p>",
      "rawMarkdown": "NN usually won't perform as well as the tree-based boosting methods like XGB, CatBoost and LGBM with small-sized tabular data (like less than 400k rows). It doesn't mean they can be competitive, you just have to experiment with a variety of architectures and hyperparameters.",
      "votes": null
    },
    {
      "id": "3036531",
      "postDate": "11/04/2024 16:37:56",
      "content": "<p><a href=\"https://www.kaggle.com/vitormeurerbesen\" target=\"_blank\">@vitormeurerbesen</a>  thank you, yes NNs nedd a large data </p>",
      "rawMarkdown": "vitormeurerbesen  thank you, yes NNs nedd a large data",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3027322,
      "author_name": "vitormeurerbesen",
      "author_url": "",
      "post_date": "10/24/2024 17:14:08",
      "content": "<p>NN usually won't perform as well as the tree-based boosting methods like XGB, CatBoost and LGBM with small-sized tabular data (like less than 400k rows). It doesn't mean they can be competitive, you just have to experiment with a variety of architectures and hyperparameters.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3036531,
          "author_name": "saidkoussi",
          "author_url": "",
          "post_date": "11/04/2024 16:37:56",
          "content": "<p><a href=\"https://www.kaggle.com/vitormeurerbesen\" target=\"_blank\">@vitormeurerbesen</a>  thank you, yes NNs nedd a large data </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3023663": "when we can prefer NNs models vs modern models for small size of tabular data?",
    "3027322": "NN usually won't perform as well as the tree-based boosting methods like XGB, CatBoost and LGBM with small-sized tabular data (like less than 400k rows). It doesn't mean they can be competitive, you just have to experiment with a variety of architectures and hyperparameters.",
    "3036531": "vitormeurerbesen  thank you, yes NNs nedd a large data"
  },
  "source": "meta"
}