{
  "id": 190291,
  "title": "How come does LGB outperform?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/190291",
  "author_name": "",
  "post_date": "2020-10-11T04:31:00.656129600Z",
  "votes": 12,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Among the three popular GBDT libraries (lightGBM, XGBoost, CatBoost), it seems that in this competition (still very early stage, though) LGB outperforms others. Maybe it's just a hyperparameter thing, but the validation score for those three models are in my case:</p>\n<ul>\n<li>LGB = 0.7588</li>\n<li>XGB = 0.6427</li>\n<li>CatB = 0.6434</li>\n</ul>\n<p>Here is my kernel to explore GBDTs in this competition.<br>\n<a href=\"https://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline\" target=\"_blank\">https://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline</a></p>\n<p>I've not seen such a huge difference in GBDT's performance before…do you have any clue?</p>",
  "messages": [
    {
      "id": "1045811",
      "postDate": "10/11/2020 04:31:00",
      "content": "<p>Among the three popular GBDT libraries (lightGBM, XGBoost, CatBoost), it seems that in this competition (still very early stage, though) LGB outperforms others. Maybe it's just a hyperparameter thing, but the validation score for those three models are in my case:</p>\n<ul>\n<li>LGB = 0.7588</li>\n<li>XGB = 0.6427</li>\n<li>CatB = 0.6434</li>\n</ul>\n<p>Here is my kernel to explore GBDTs in this competition.<br>\n<a href=\"https://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline\" target=\"_blank\">https://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline</a></p>\n<p>I've not seen such a huge difference in GBDT's performance before…do you have any clue?</p>",
      "rawMarkdown": "Among the three popular GBDT libraries (lightGBM, XGBoost, CatBoost), it seems that in this competition (still very early stage, though) LGB outperforms others. Maybe it's just a hyperparameter thing, but the validation score for those three models are in my case:\n\n- LGB = 0.7588\n- XGB = 0.6427\n- CatB = 0.6434\n\nHere is my kernel to explore GBDTs in this competition.\nhttps://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline\n\nI've not seen such a huge difference in GBDT's performance before...do you have any clue?",
      "votes": null
    },
    {
      "id": "1046037",
      "postDate": "10/11/2020 09:41:36",
      "content": "<p>Of course should be a hp issue and probably one reason would be the no. of training samples. I have seen before that LGBM outperforms XGB with 'very large' datasets. Thanks for sharing your notebook btw.</p>",
      "rawMarkdown": "Of course should be a hp issue and probably one reason would be the no. of training samples. I have seen before that LGBM outperforms XGB with 'very large' datasets. Thanks for sharing your notebook btw.",
      "votes": null
    },
    {
      "id": "1046904",
      "postDate": "10/12/2020 04:46:41",
      "content": "<p>Thanks for sharing</p>",
      "rawMarkdown": "Thanks for sharing",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1046037,
      "author_name": "imeintanis",
      "author_url": "",
      "post_date": "10/11/2020 09:41:36",
      "content": "<p>Of course should be a hp issue and probably one reason would be the no. of training samples. I have seen before that LGBM outperforms XGB with 'very large' datasets. Thanks for sharing your notebook btw.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1046904,
      "author_name": "kushal1506",
      "author_url": "",
      "post_date": "10/12/2020 04:46:41",
      "content": "<p>Thanks for sharing</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1045811": "Among the three popular GBDT libraries (lightGBM, XGBoost, CatBoost), it seems that in this competition (still very early stage, though) LGB outperforms others. Maybe it's just a hyperparameter thing, but the validation score for those three models are in my case:\n\n- LGB = 0.7588\n- XGB = 0.6427\n- CatB = 0.6434\n\nHere is my kernel to explore GBDTs in this competition.\nhttps://www.kaggle.com/code1110/riiid-gbdt-pipeline-baseline\n\nI've not seen such a huge difference in GBDT's performance before...do you have any clue?",
    "1046037": "Of course should be a hp issue and probably one reason would be the no. of training samples. I have seen before that LGBM outperforms XGB with 'very large' datasets. Thanks for sharing your notebook btw.",
    "1046904": "Thanks for sharing"
  },
  "source": "meta"
}