{
  "id": 76478,
  "title": "local score much higher than LB",
  "url": "/competitions/quora-insincere-questions-classification/discussion/76478",
  "author_name": "",
  "post_date": "2019-01-03T08:34:08.261181600Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>For example:\nlocal score 0.695, LB 0.655; \nlocal 0.692, LB 0.652;\nlocal 0.69, LB 0.66; \nlocal 0.68, LB 0.659.\nIt seems that maybe I overfit the validation, but I checked all my code many times and I find nothing wrong. \nI also cannot tell how the local relate to LB. Does anyone have the same problem?</p>",
  "messages": [
    {
      "id": "449498",
      "postDate": "01/03/2019 08:34:08",
      "content": "<p>For example:\nlocal score 0.695, LB 0.655; \nlocal 0.692, LB 0.652;\nlocal 0.69, LB 0.66; \nlocal 0.68, LB 0.659.\nIt seems that maybe I overfit the validation, but I checked all my code many times and I find nothing wrong. \nI also cannot tell how the local relate to LB. Does anyone have the same problem?</p>",
      "rawMarkdown": "For example:\nlocal score 0.695, LB 0.655; \nlocal 0.692, LB 0.652;\nlocal 0.69, LB 0.66; \nlocal 0.68, LB 0.659.\nIt seems that maybe I overfit the validation, but I checked all my code many times and I find nothing wrong. \nI also cannot tell how the local relate to LB. Does anyone have the same problem?",
      "votes": null
    },
    {
      "id": "449546",
      "postDate": "01/03/2019 10:33:40",
      "content": "<p>I have higher validation scores too, but the difference is not too dramatic. Seems that you have some target leakages in your pipeline. </p>",
      "rawMarkdown": "I have higher validation scores too, but the difference is not too dramatic. Seems that you have some target leakages in your pipeline.",
      "votes": null
    },
    {
      "id": "449594",
      "postDate": "01/03/2019 12:08:55",
      "content": "<p>target leakages ,check your model \"Object\" is new NN() in kfold's for loop</p>",
      "rawMarkdown": "target leakages ,check your model \"Object\" is new NN() in kfold's for loop",
      "votes": null
    },
    {
      "id": "450887",
      "postDate": "01/06/2019 02:24:09",
      "content": "<p>I notice that you do not mention CV.  Do you use cross validation to get local score?  My cv score is about 0.01 ~ 0.02 higher than LB (single model, no fold averge), but the gap shouldn't be that large.  </p>",
      "rawMarkdown": "I notice that you do not mention CV.  Do you use cross validation to get local score?  My cv score is about 0.01 ~ 0.02 higher than LB (single model, no fold averge), but the gap shouldn't be that large.",
      "votes": null
    },
    {
      "id": "451056",
      "postDate": "01/06/2019 11:09:24",
      "content": "<p>Yes, finally, I find out I have target leakages. Thank you so much.</p>",
      "rawMarkdown": "Yes, finally, I find out I have target leakages. Thank you so much.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 449546,
      "author_name": "",
      "author_url": "",
      "post_date": "01/03/2019 10:33:40",
      "content": "<p>I have higher validation scores too, but the difference is not too dramatic. Seems that you have some target leakages in your pipeline. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 449594,
      "author_name": "baomengjiao",
      "author_url": "",
      "post_date": "01/03/2019 12:08:55",
      "content": "<p>target leakages ,check your model \"Object\" is new NN() in kfold's for loop</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 450887,
      "author_name": "jiangm",
      "author_url": "",
      "post_date": "01/06/2019 02:24:09",
      "content": "<p>I notice that you do not mention CV.  Do you use cross validation to get local score?  My cv score is about 0.01 ~ 0.02 higher than LB (single model, no fold averge), but the gap shouldn't be that large.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 451056,
      "author_name": "wxytalent",
      "author_url": "",
      "post_date": "01/06/2019 11:09:24",
      "content": "<p>Yes, finally, I find out I have target leakages. Thank you so much.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "449498": "For example:\nlocal score 0.695, LB 0.655; \nlocal 0.692, LB 0.652;\nlocal 0.69, LB 0.66; \nlocal 0.68, LB 0.659.\nIt seems that maybe I overfit the validation, but I checked all my code many times and I find nothing wrong. \nI also cannot tell how the local relate to LB. Does anyone have the same problem?",
    "449546": "I have higher validation scores too, but the difference is not too dramatic. Seems that you have some target leakages in your pipeline.",
    "449594": "target leakages ,check your model \"Object\" is new NN() in kfold's for loop",
    "450887": "I notice that you do not mention CV.  Do you use cross validation to get local score?  My cv score is about 0.01 ~ 0.02 higher than LB (single model, no fold averge), but the gap shouldn't be that large.",
    "451056": "Yes, finally, I find out I have target leakages. Thank you so much."
  },
  "source": "meta"
}