{
  "id": 92028,
  "title": "Why better CV get bad LB?",
  "url": "/competitions/LANL-Earthquake-Prediction/discussion/92028",
  "author_name": "",
  "post_date": "2019-05-12T09:01:20.749930300Z",
  "votes": 1,
  "comment_count": 5,
  "views": 0,
  "content": "<p>previous version:  lgb get→ CV:2.08473 LB:1.457\nnow version: lgb+something change get→ CV:2.0318 LB:1.467\nPlease tell me if I miss something.</p>",
  "messages": [
    {
      "id": "530245",
      "postDate": "05/12/2019 09:01:20",
      "content": "<p>previous version:  lgb get→ CV:2.08473 LB:1.457\nnow version: lgb+something change get→ CV:2.0318 LB:1.467\nPlease tell me if I miss something.</p>",
      "rawMarkdown": "previous version:  lgb get→ CV:2.08473 LB:1.457\nnow version: lgb+something change get→ CV:2.0318 LB:1.467\nPlease tell me if I miss something.",
      "votes": null
    },
    {
      "id": "530266",
      "postDate": "05/12/2019 10:18:48",
      "content": "<p>Could be standard deviation. Look at the LB score as just another fold. </p>\n\n<p>Also, have a look at the oof file of your model. Does it differ a lot from the last. There might be your explanation. </p>",
      "rawMarkdown": "Could be standard deviation. Look at the LB score as just another fold. \n\nAlso, have a look at the oof file of your model. Does it differ a lot from the last. There might be your explanation.",
      "votes": null
    },
    {
      "id": "530272",
      "postDate": "05/12/2019 10:41:07",
      "content": "<p>Thanks for your reply. I change OOF in new version,believe there is the problem.</p>",
      "rawMarkdown": "Thanks for your reply. I change OOF in new version,believe there is the problem.",
      "votes": null
    },
    {
      "id": "530340",
      "postDate": "05/12/2019 14:36:54",
      "content": "<p>-Public LB is only 341 segment. Train data has 4000+ sample.\n-Distribution between public LB and Train data might also be very different.\n-Are you sure you are using a reliable CV in a first place ?</p>",
      "rawMarkdown": "Public LB is only 341 segment. Train data has 4000+ sample.\n-Distribution between public LB and Train data might also be very different.\n-Are you sure you are using a reliable CV in a first place ?",
      "votes": null
    },
    {
      "id": "530366",
      "postDate": "05/12/2019 16:22:48",
      "content": "<p>I will check it  in tomorrow. \nMy opinion is trust my cv.\n   1.Becasue  there is only  341 in LB.\n   2.I only improve a little basing on Andrews baseline,which is  useful in another kernel.</p>",
      "rawMarkdown": "I will check it  in tomorrow. \nMy opinion is trust my cv.\n   1.Becasue  there is only  341 in LB.\n   2.I only improve a little basing on Andrews baseline,which is  useful in another kernel.",
      "votes": null
    },
    {
      "id": "530713",
      "postDate": "05/13/2019 13:51:39",
      "content": "<p>I can't find reason about this question,but maybe it is not the key in this competition,trust cv</p>",
      "rawMarkdown": "I can't find reason about this question,but maybe it is not the key in this competition,trust cv",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 530266,
      "author_name": "ilu000",
      "author_url": "",
      "post_date": "05/12/2019 10:18:48",
      "content": "<p>Could be standard deviation. Look at the LB score as just another fold. </p>\n\n<p>Also, have a look at the oof file of your model. Does it differ a lot from the last. There might be your explanation. </p>",
      "votes": null,
      "replies": [
        {
          "id": 530272,
          "author_name": "haonanyao",
          "author_url": "",
          "post_date": "05/12/2019 10:41:07",
          "content": "<p>Thanks for your reply. I change OOF in new version,believe there is the problem.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 530340,
      "author_name": "areveillon",
      "author_url": "",
      "post_date": "05/12/2019 14:36:54",
      "content": "<p>-Public LB is only 341 segment. Train data has 4000+ sample.\n-Distribution between public LB and Train data might also be very different.\n-Are you sure you are using a reliable CV in a first place ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 530366,
          "author_name": "haonanyao",
          "author_url": "",
          "post_date": "05/12/2019 16:22:48",
          "content": "<p>I will check it  in tomorrow. \nMy opinion is trust my cv.\n   1.Becasue  there is only  341 in LB.\n   2.I only improve a little basing on Andrews baseline,which is  useful in another kernel.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 530713,
          "author_name": "haonanyao",
          "author_url": "",
          "post_date": "05/13/2019 13:51:39",
          "content": "<p>I can't find reason about this question,but maybe it is not the key in this competition,trust cv</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "530245": "previous version:  lgb get→ CV:2.08473 LB:1.457\nnow version: lgb+something change get→ CV:2.0318 LB:1.467\nPlease tell me if I miss something.",
    "530266": "Could be standard deviation. Look at the LB score as just another fold. \n\nAlso, have a look at the oof file of your model. Does it differ a lot from the last. There might be your explanation.",
    "530272": "Thanks for your reply. I change OOF in new version,believe there is the problem.",
    "530340": "Public LB is only 341 segment. Train data has 4000+ sample.\n-Distribution between public LB and Train data might also be very different.\n-Are you sure you are using a reliable CV in a first place ?",
    "530366": "I will check it  in tomorrow. \nMy opinion is trust my cv.\n   1.Becasue  there is only  341 in LB.\n   2.I only improve a little basing on Andrews baseline,which is  useful in another kernel.",
    "530713": "I can't find reason about this question,but maybe it is not the key in this competition,trust cv"
  },
  "source": "meta"
}