{
  "id": 99870,
  "title": "LightGBM stopping tolerance",
  "url": "/competitions/champs-scalar-coupling/discussion/99870",
  "author_name": "",
  "post_date": "2019-07-15T01:16:08.859238300Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Is there a way to specify a stopping tolerance value in lightgbm in addition to the number of stopping rounds? In some of my experiments, I notice only a fourth or fifth digit improvement with the increase in boosting rounds, and so I was wondering if having a stopping tolerance value like 10^-3 might stop the training earlier.</p>",
  "messages": [
    {
      "id": "575049",
      "postDate": "07/15/2019 01:16:08",
      "content": "<p>Is there a way to specify a stopping tolerance value in lightgbm in addition to the number of stopping rounds? In some of my experiments, I notice only a fourth or fifth digit improvement with the increase in boosting rounds, and so I was wondering if having a stopping tolerance value like 10^-3 might stop the training earlier.</p>",
      "rawMarkdown": "Is there a way to specify a stopping tolerance value in lightgbm in addition to the number of stopping rounds? In some of my experiments, I notice only a fourth or fifth digit improvement with the increase in boosting rounds, and so I was wondering if having a stopping tolerance value like 10^-3 might stop the training earlier.",
      "votes": null
    },
    {
      "id": "575115",
      "postDate": "07/15/2019 04:37:02",
      "content": "<p>Good question - I have wondered that myself.  Hope someone has a good answer.</p>",
      "rawMarkdown": "Good question - I have wondered that myself.  Hope someone has a good answer.",
      "votes": null
    },
    {
      "id": "575351",
      "postDate": "07/15/2019 10:25:51",
      "content": "<p>I think there are no parameter or option for achieving it in lightgbm library.\nHowever using callback function makes that early stopping possible.</p>\n\n<p>This is a kernel which customizes the early stopping condition by callback function. (Note: It is just an example of customizing early stopping, not the stopping condition you hope)\n<a href=\"https://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap\">https://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap</a></p>\n\n<p>If you make the callback function like it, it will be possible!!</p>",
      "rawMarkdown": "I think there are no parameter or option for achieving it in lightgbm library.\nHowever using callback function makes that early stopping possible.\n\nThis is a kernel which customizes the early stopping condition by callback function. (Note: It is just an example of customizing early stopping, not the stopping condition you hope)\nhttps://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap\n\nIf you make the callback function like it, it will be possible!!",
      "votes": null
    },
    {
      "id": "576747",
      "postDate": "07/16/2019 00:52:57",
      "content": "<p>Thank you very much. This might be what I am looking for.</p>",
      "rawMarkdown": "Thank you very much. This might be what I am looking for.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 575115,
      "author_name": "pcjimmmy",
      "author_url": "",
      "post_date": "07/15/2019 04:37:02",
      "content": "<p>Good question - I have wondered that myself.  Hope someone has a good answer.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 575351,
      "author_name": "go5kuramubon",
      "author_url": "",
      "post_date": "07/15/2019 10:25:51",
      "content": "<p>I think there are no parameter or option for achieving it in lightgbm library.\nHowever using callback function makes that early stopping possible.</p>\n\n<p>This is a kernel which customizes the early stopping condition by callback function. (Note: It is just an example of customizing early stopping, not the stopping condition you hope)\n<a href=\"https://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap\">https://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap</a></p>\n\n<p>If you make the callback function like it, it will be possible!!</p>",
      "votes": null,
      "replies": [
        {
          "id": 576747,
          "author_name": "hidasri",
          "author_url": "",
          "post_date": "07/16/2019 00:52:57",
          "content": "<p>Thank you very much. This might be what I am looking for.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "575049": "Is there a way to specify a stopping tolerance value in lightgbm in addition to the number of stopping rounds? In some of my experiments, I notice only a fourth or fifth digit improvement with the increase in boosting rounds, and so I was wondering if having a stopping tolerance value like 10^-3 might stop the training earlier.",
    "575115": "Good question - I have wondered that myself.  Hope someone has a good answer.",
    "575351": "I think there are no parameter or option for achieving it in lightgbm library.\nHowever using callback function makes that early stopping possible.\n\nThis is a kernel which customizes the early stopping condition by callback function. (Note: It is just an example of customizing early stopping, not the stopping condition you hope)\nhttps://www.kaggle.com/alexfir/lgbm-early-stopping-based-on-train-validation-gap\n\nIf you make the callback function like it, it will be possible!!",
    "576747": "Thank you very much. This might be what I am looking for."
  },
  "source": "meta"
}