{
  "id": 207773,
  "title": "Does early stopping work on LightGBM?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/207773",
  "author_name": "",
  "post_date": "2020-12-31T07:26:49.011217400Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>My valid score(AUC) keeps increasing by 0.0000x in every epoch.</p>\n<p>At the first phase of training, training's auroc is a little less than valid's, but as time goes, it's getting bigger and finally surpass the valid's one and their gap is also getting larger.</p>\n<p>When do you guys stop training?</p>",
  "messages": [
    {
      "id": "1133403",
      "postDate": "12/31/2020 07:26:49",
      "content": "<p>My valid score(AUC) keeps increasing by 0.0000x in every epoch.</p>\n<p>At the first phase of training, training's auroc is a little less than valid's, but as time goes, it's getting bigger and finally surpass the valid's one and their gap is also getting larger.</p>\n<p>When do you guys stop training?</p>",
      "rawMarkdown": "My valid score(AUC) keeps increasing by 0.0000x in every epoch.\n\nAt the first phase of training, training's auroc is a little less than valid's, but as time goes, it's getting bigger and finally surpass the valid's one and their gap is also getting larger.\n\nWhen do you guys stop training?",
      "votes": null
    },
    {
      "id": "1133519",
      "postDate": "12/31/2020 09:28:33",
      "content": "<p>I usually set a small learning rate &lt; 0.05, and use more than one validation set, and stop training when ROC validation start worsening. My LB and CV are pretty consistent even though the training ROC is &gt;&gt; validation ROC.</p>",
      "rawMarkdown": "I usually set a small learning rate < 0.05, and use more than one validation set, and stop training when ROC validation start worsening. My LB and CV are pretty consistent even though the training ROC is >> validation ROC.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1133519,
      "author_name": "abdessalemboukil",
      "author_url": "",
      "post_date": "12/31/2020 09:28:33",
      "content": "<p>I usually set a small learning rate &lt; 0.05, and use more than one validation set, and stop training when ROC validation start worsening. My LB and CV are pretty consistent even though the training ROC is &gt;&gt; validation ROC.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1133403": "My valid score(AUC) keeps increasing by 0.0000x in every epoch.\n\nAt the first phase of training, training's auroc is a little less than valid's, but as time goes, it's getting bigger and finally surpass the valid's one and their gap is also getting larger.\n\nWhen do you guys stop training?",
    "1133519": "I usually set a small learning rate < 0.05, and use more than one validation set, and stop training when ROC validation start worsening. My LB and CV are pretty consistent even though the training ROC is >> validation ROC."
  },
  "source": "meta"
}