{
  "id": 73762,
  "title": "why LightGBM?",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/73762",
  "author_name": "richarde",
  "post_date": "2018-12-05T12:41:26.572000",
  "votes": 3,
  "comment_count": 12,
  "views": 0,
  "content": "<p>Most kernels are using lightgbm rather than xgboost.  Is there a technical reason for this? </p>",
  "messages": [
    {
      "id": 433743,
      "postDate": "2018-12-05T12:41:26.573Z",
      "content": "<p>Most kernels are using lightgbm rather than xgboost.  Is there a technical reason for this? </p>",
      "rawMarkdown": "Most kernels are using lightgbm rather than xgboost.  Is there a technical reason for this? ",
      "votes": 3
    },
    {
      "id": 433938,
      "postDate": "2018-12-05T17:13:19.470Z",
      "content": "<ol>\n<li>Fast speed</li>\n<li>Handles categorical features if we input the names of features into the model.</li>\n<li>It internally handles annoying null data.</li>\n</ol>\n\n<p>You can read more information of lightgbm here. <a href=\"https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html\">https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html</a></p>",
      "rawMarkdown": "1. Fast speed\n2. Handles categorical features if we input the names of features into the model.\n3. It internally handles annoying null data.\n\nYou can read more information of lightgbm here. https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html",
      "votes": 2,
      "replies": [
        {
          "id": 433989,
          "postDate": "2018-12-05T18:30:16.393Z",
          "content": "<p>LGBM document site is very usefull thanx</p>",
          "rawMarkdown": "LGBM document site is very usefull thanx"
        }
      ]
    },
    {
      "id": 433750,
      "postDate": "2018-12-05T12:46:55.643Z",
      "content": "<p>lightgbm is much faster and often has comparable or higher quality</p>",
      "rawMarkdown": "lightgbm is much faster and often has comparable or higher quality",
      "votes": 2,
      "replies": [
        {
          "id": 433754,
          "postDate": "2018-12-05T12:48:54.287Z",
          "content": "<p>So there is nothing about the dataset in particular? lightgbm is now the goto algorithm in shallow machine learning?</p>",
          "rawMarkdown": "So there is nothing about the dataset in particular? lightgbm is now the goto algorithm in shallow machine learning?",
          "votes": 4
        },
        {
          "id": 433779,
          "postDate": "2018-12-05T13:20:19.850Z",
          "content": "<p>Yes, but in general LGB is one of the best models for tabular data. At least currently.</p>",
          "rawMarkdown": "Yes, but in general LGB is one of the best models for tabular data. At least currently.",
          "votes": 1
        }
      ]
    },
    {
      "id": 460455,
      "postDate": "2019-01-23T18:27:46.037Z",
      "content": "<p>Algorithms wise, are lightgbm the same as xgboost?</p>",
      "rawMarkdown": "Algorithms wise, are lightgbm the same as xgboost?"
    },
    {
      "id": 433868,
      "postDate": "2018-12-05T15:22:33.007Z",
      "content": "<p>Mostly speed (in my case at least)  - xgboost gives slightly higher quality results in some situations, but is slower.</p>",
      "rawMarkdown": "Mostly speed (in my case at least)  - xgboost gives slightly higher quality results in some situations, but is slower.",
      "replies": [
        {
          "id": 434610,
          "postDate": "2018-12-06T17:11:39.813Z",
          "content": "<p>@Konrad </p>\n\n<p>If you read carefully the xgboost documentation, you'll find tips to make it run much faster than LightGBM.</p>\n\n<p><a href=\"https://xgboost.readthedocs.io/en/latest/parameter.html\">https://xgboost.readthedocs.io/en/latest/parameter.html</a></p>",
          "rawMarkdown": "@Konrad \n\nIf you read carefully the xgboost documentation, you'll find tips to make it run much faster than LightGBM.\n\nhttps://xgboost.readthedocs.io/en/latest/parameter.html"
        },
        {
          "id": 463954,
          "postDate": "2019-01-31T00:18:54.107Z",
          "content": "<p>my Xgboost model is way worse ng lightgbm</p>",
          "rawMarkdown": "my Xgboost model is way worse ng lightgbm"
        }
      ]
    },
    {
      "id": 433757,
      "postDate": "2018-12-05T12:50:34.270Z",
      "content": "<p>Computation time, lower memory usage and usually better accuracy I guess.</p>",
      "rawMarkdown": "Computation time, lower memory usage and usually better accuracy I guess."
    },
    {
      "id": 433951,
      "postDate": "2018-12-05T17:26:25.733Z",
      "rawMarkdown": "",
      "isDeleted": true,
      "replies": [
        {
          "id": 434605,
          "postDate": "2018-12-06T17:08:24.463Z",
          "content": "<p>@Soufiane </p>\n\n<p>You can use LightGBM with <strong>GPU</strong> too adding line :  'device' : 'gpu' in your params.</p>",
          "rawMarkdown": "@Soufiane \n\nYou can use LightGBM with **GPU** too adding line :  'device' : 'gpu' in your params."
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 433938,
      "author_name": "Youhan Lee",
      "author_url": "",
      "post_date": "2018-12-05T17:13:19.470000",
      "content": "<ol>\n<li>Fast speed</li>\n<li>Handles categorical features if we input the names of features into the model.</li>\n<li>It internally handles annoying null data.</li>\n</ol>\n\n<p>You can read more information of lightgbm here. <a href=\"https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html\">https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html</a></p>",
      "votes": 2,
      "replies": [
        {
          "id": 433989,
          "author_name": "mahmut haktan",
          "author_url": "",
          "post_date": "2018-12-05T18:30:16.393000",
          "content": "<p>LGBM document site is very usefull thanx</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 433750,
      "author_name": "Andrey Lukyanenko",
      "author_url": "",
      "post_date": "2018-12-05T12:46:55.643000",
      "content": "<p>lightgbm is much faster and often has comparable or higher quality</p>",
      "votes": 2,
      "replies": [
        {
          "id": 433754,
          "author_name": "richarde",
          "author_url": "",
          "post_date": "2018-12-05T12:48:54.287000",
          "content": "<p>So there is nothing about the dataset in particular? lightgbm is now the goto algorithm in shallow machine learning?</p>",
          "votes": 4,
          "replies": []
        },
        {
          "id": 433779,
          "author_name": "Andrey Lukyanenko",
          "author_url": "",
          "post_date": "2018-12-05T13:20:19.850000",
          "content": "<p>Yes, but in general LGB is one of the best models for tabular data. At least currently.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 460455,
      "author_name": "LiwenHuang",
      "author_url": "",
      "post_date": "2019-01-23T18:27:46.037000",
      "content": "<p>Algorithms wise, are lightgbm the same as xgboost?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 433868,
      "author_name": "Konrad Banachewicz",
      "author_url": "",
      "post_date": "2018-12-05T15:22:33.007000",
      "content": "<p>Mostly speed (in my case at least)  - xgboost gives slightly higher quality results in some situations, but is slower.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 434610,
          "author_name": "mezoganet",
          "author_url": "",
          "post_date": "2018-12-06T17:11:39.813000",
          "content": "<p>@Konrad </p>\n\n<p>If you read carefully the xgboost documentation, you'll find tips to make it run much faster than LightGBM.</p>\n\n<p><a href=\"https://xgboost.readthedocs.io/en/latest/parameter.html\">https://xgboost.readthedocs.io/en/latest/parameter.html</a></p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 463954,
          "author_name": "dylonLL",
          "author_url": "",
          "post_date": "2019-01-31T00:18:54.107000",
          "content": "<p>my Xgboost model is way worse ng lightgbm</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 433757,
      "author_name": "Rafał Baka",
      "author_url": "",
      "post_date": "2018-12-05T12:50:34.270000",
      "content": "<p>Computation time, lower memory usage and usually better accuracy I guess.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 433951,
      "author_name": "",
      "author_url": "",
      "post_date": "2018-12-05T17:26:25.733000",
      "content": "",
      "votes": 0,
      "replies": [
        {
          "id": 434605,
          "author_name": "mezoganet",
          "author_url": "",
          "post_date": "2018-12-06T17:08:24.463000",
          "content": "<p>@Soufiane </p>\n\n<p>You can use LightGBM with <strong>GPU</strong> too adding line :  'device' : 'gpu' in your params.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "433743": "Most kernels are using lightgbm rather than xgboost.  Is there a technical reason for this? ",
    "433938": "1. Fast speed\n2. Handles categorical features if we input the names of features into the model.\n3. It internally handles annoying null data.\n\nYou can read more information of lightgbm here. https://lightgbm.readthedocs.io/en/latest/Advanced-Topics.html",
    "433750": "lightgbm is much faster and often has comparable or higher quality",
    "460455": "Algorithms wise, are lightgbm the same as xgboost?",
    "433868": "Mostly speed (in my case at least)  - xgboost gives slightly higher quality results in some situations, but is slower.",
    "433757": "Computation time, lower memory usage and usually better accuracy I guess.",
    "433951": ""
  }
}