{
  "id": 107721,
  "title": "LB better then training set",
  "url": "/competitions/understanding_cloud_organization/discussion/107721",
  "author_name": "",
  "post_date": "2019-09-06T06:47:30.628419300Z",
  "votes": 1,
  "comment_count": 8,
  "views": 0,
  "content": "<p>My LB 0.58 training set .52 validation- .5 . What is the reason,what do you think ?</p>",
  "messages": [
    {
      "id": "619396",
      "postDate": "09/06/2019 06:47:30",
      "content": "<p>My LB 0.58 training set .52 validation- .5 . What is the reason,what do you think ?</p>",
      "rawMarkdown": "My LB 0.58 training set .52 validation- .5 . What is the reason,what do you think ?",
      "votes": null
    },
    {
      "id": "619468",
      "postDate": "09/06/2019 08:11:04",
      "content": "<p><code>The Dice coefficient is defined to be 1 when both X and Y are empty.</code> Have you take this into account when calculating the validation score?</p>",
      "rawMarkdown": "`The Dice coefficient is defined to be 1 when both X and Y are empty. ` Have you take this into account when calculating the validation score?",
      "votes": null
    },
    {
      "id": "619714",
      "postDate": "09/06/2019 13:56:42",
      "content": "<p>thank you for question I need to think</p>",
      "rawMarkdown": "thank you for question I need to think",
      "votes": null
    },
    {
      "id": "620766",
      "postDate": "09/08/2019 00:21:55",
      "content": "<p>Don't Trust Lb.i trusted LB and was thrown away from 6% to 99% for huge shake up in aptos blindness detection competition,always trust your CV</p>",
      "rawMarkdown": "Don't Trust Lb.i trusted LB and was thrown away from 6% to 99% for huge shake up in aptos blindness detection competition,always trust your CV",
      "votes": null
    },
    {
      "id": "620774",
      "postDate": "09/08/2019 00:31:20",
      "content": "<p>What happened there and so many people got 0 on private LB? Is it because of the running error or CV shakeup?</p>",
      "rawMarkdown": "What happened there and so many people got 0 on private LB? Is it because of the running error or CV shakeup?",
      "votes": null
    },
    {
      "id": "621136",
      "postDate": "09/08/2019 08:36:06",
      "content": "<p>Almost all of our submission got 0.00 private lb,but we had few models that worked well in public lb,,our 1 submission had 0.815 public lb score,,,,,,the one with 0.68 public lb score had more than 90% cross validation score but it is still 0.00 in private lb,,,experienced terrible competition mate,i won't spend much time on any kernel competition ever,i don't trust the evaluation techniques of such competitions</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fa8874e29fe3a73b97809a42a6147ed54%2Fzero.png?generation=1567931601576225&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Almost all of our submission got 0.00 private lb,but we had few models that worked well in public lb,,our 1 submission had 0.815 public lb score,,,,,,the one with 0.68 public lb score had more than 90% cross validation score but it is still 0.00 in private lb,,,experienced terrible competition mate,i won't spend much time on any kernel competition ever,i don't trust the evaluation techniques of such competitions\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fa8874e29fe3a73b97809a42a6147ed54%2Fzero.png?generation=1567931601576225&amp;alt=media)",
      "votes": null
    },
    {
      "id": "621157",
      "postDate": "09/08/2019 09:03:43",
      "content": "<p>My experiencw in Jigsaw was similar. Only one submission was valid in that comp finally...</p>",
      "rawMarkdown": "My experiencw in Jigsaw was similar. Only one submission was valid in that comp finally...",
      "votes": null
    },
    {
      "id": "621160",
      "postDate": "09/08/2019 09:07:22",
      "content": "<p>now i understand what went wrong...i selected fast submission kernels accidentally and for that reason i got 0.00 lb</p>",
      "rawMarkdown": "now i understand what went wrong...i selected fast submission kernels accidentally and for that reason i got 0.00 lb",
      "votes": null
    },
    {
      "id": "621891",
      "postDate": "09/09/2019 04:36:26",
      "content": "<p>Very well possible that the data presented in public is \"easier\" than training data. As stressed by others, try to get a good validation strategy and it should give a clearer indication of model performance. e.g. 5 k-fold CV can be much clearer to check than a single split.</p>",
      "rawMarkdown": "Very well possible that the data presented in public is \"easier\" than training data. As stressed by others, try to get a good validation strategy and it should give a clearer indication of model performance. e.g. 5 k-fold CV can be much clearer to check than a single split.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 619468,
      "author_name": "gogo827jz",
      "author_url": "",
      "post_date": "09/06/2019 08:11:04",
      "content": "<p><code>The Dice coefficient is defined to be 1 when both X and Y are empty.</code> Have you take this into account when calculating the validation score?</p>",
      "votes": null,
      "replies": [
        {
          "id": 619714,
          "author_name": "jonasmatuzas",
          "author_url": "",
          "post_date": "09/06/2019 13:56:42",
          "content": "<p>thank you for question I need to think</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 620766,
      "author_name": "mobassir",
      "author_url": "",
      "post_date": "09/08/2019 00:21:55",
      "content": "<p>Don't Trust Lb.i trusted LB and was thrown away from 6% to 99% for huge shake up in aptos blindness detection competition,always trust your CV</p>",
      "votes": null,
      "replies": [
        {
          "id": 620774,
          "author_name": "gogo827jz",
          "author_url": "",
          "post_date": "09/08/2019 00:31:20",
          "content": "<p>What happened there and so many people got 0 on private LB? Is it because of the running error or CV shakeup?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 621136,
          "author_name": "mobassir",
          "author_url": "",
          "post_date": "09/08/2019 08:36:06",
          "content": "<p>Almost all of our submission got 0.00 private lb,but we had few models that worked well in public lb,,our 1 submission had 0.815 public lb score,,,,,,the one with 0.68 public lb score had more than 90% cross validation score but it is still 0.00 in private lb,,,experienced terrible competition mate,i won't spend much time on any kernel competition ever,i don't trust the evaluation techniques of such competitions</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fa8874e29fe3a73b97809a42a6147ed54%2Fzero.png?generation=1567931601576225&amp;alt=media\" alt=\"\"></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 621157,
          "author_name": "gogo827jz",
          "author_url": "",
          "post_date": "09/08/2019 09:03:43",
          "content": "<p>My experiencw in Jigsaw was similar. Only one submission was valid in that comp finally...</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 621160,
          "author_name": "mobassir",
          "author_url": "",
          "post_date": "09/08/2019 09:07:22",
          "content": "<p>now i understand what went wrong...i selected fast submission kernels accidentally and for that reason i got 0.00 lb</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 621891,
      "author_name": "joonl04",
      "author_url": "",
      "post_date": "09/09/2019 04:36:26",
      "content": "<p>Very well possible that the data presented in public is \"easier\" than training data. As stressed by others, try to get a good validation strategy and it should give a clearer indication of model performance. e.g. 5 k-fold CV can be much clearer to check than a single split.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "619396": "My LB 0.58 training set .52 validation- .5 . What is the reason,what do you think ?",
    "619468": "`The Dice coefficient is defined to be 1 when both X and Y are empty. ` Have you take this into account when calculating the validation score?",
    "619714": "thank you for question I need to think",
    "620766": "Don't Trust Lb.i trusted LB and was thrown away from 6% to 99% for huge shake up in aptos blindness detection competition,always trust your CV",
    "620774": "What happened there and so many people got 0 on private LB? Is it because of the running error or CV shakeup?",
    "621136": "Almost all of our submission got 0.00 private lb,but we had few models that worked well in public lb,,our 1 submission had 0.815 public lb score,,,,,,the one with 0.68 public lb score had more than 90% cross validation score but it is still 0.00 in private lb,,,experienced terrible competition mate,i won't spend much time on any kernel competition ever,i don't trust the evaluation techniques of such competitions\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fa8874e29fe3a73b97809a42a6147ed54%2Fzero.png?generation=1567931601576225&amp;alt=media)",
    "621157": "My experiencw in Jigsaw was similar. Only one submission was valid in that comp finally...",
    "621160": "now i understand what went wrong...i selected fast submission kernels accidentally and for that reason i got 0.00 lb",
    "621891": "Very well possible that the data presented in public is \"easier\" than training data. As stressed by others, try to get a good validation strategy and it should give a clearer indication of model performance. e.g. 5 k-fold CV can be much clearer to check than a single split."
  },
  "source": "meta"
}