{
  "id": 78187,
  "title": "Hello, Do you believe CV or LB???",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/78187",
  "author_name": "",
  "post_date": "2019-01-21T00:05:41.289228500Z",
  "votes": 3,
  "comment_count": 8,
  "views": 0,
  "content": "<p>9-fold LGB only - aggregation lr = 0.015</p>\n\n<p>I try to feature engineering and check to score. \nVersion1 : CV 3.6507 LB 3.688\nVersion2 :CV 3.6506 LB 3.689\nVersion3 :CV 3.6498 LB 3.689\nVersion4 :CV 3.6481 LB 3.690\nVersion5 :CV 3.6486 LB 3.693</p>\n\n<p>Even if my <code>CV is improved</code>, <code>LB is bad</code>. how should i do it..?</p>",
  "messages": [
    {
      "id": "458979",
      "postDate": "01/21/2019 00:05:41",
      "content": "<p>9-fold LGB only - aggregation lr = 0.015</p>\n\n<p>I try to feature engineering and check to score. \nVersion1 : CV 3.6507 LB 3.688\nVersion2 :CV 3.6506 LB 3.689\nVersion3 :CV 3.6498 LB 3.689\nVersion4 :CV 3.6481 LB 3.690\nVersion5 :CV 3.6486 LB 3.693</p>\n\n<p>Even if my <code>CV is improved</code>, <code>LB is bad</code>. how should i do it..?</p>",
      "rawMarkdown": "9-fold LGB only - aggregation lr = 0.015\n\nI try to feature engineering and check to score. \nVersion1 : CV 3.6507 LB 3.688\nVersion2 :CV 3.6506 LB 3.689\nVersion3 :CV 3.6498 LB 3.689\nVersion4 :CV 3.6481 LB 3.690\nVersion5 :CV 3.6486 LB 3.693\n\nEven if my `CV is improved`, `LB is bad`. how should i do it..?",
      "votes": null
    },
    {
      "id": "458983",
      "postDate": "01/21/2019 00:17:52",
      "content": "<p>As a rational data scientist we tend to believe CV but it's really hard to say</p>\n\n<p><a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295</a></p>",
      "rawMarkdown": "As a rational data scientist we tend to believe CV but it's really hard to say\n\nhttps://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295",
      "votes": null
    },
    {
      "id": "458992",
      "postDate": "01/21/2019 01:03:22",
      "content": "<p>haha... thanks link and comment. i afraid that my model overfitting in the LB or Overfitting in the validation :(</p>",
      "rawMarkdown": "haha... thanks link and comment. i afraid that my model overfitting in the LB or Overfitting in the validation :(",
      "votes": null
    },
    {
      "id": "459129",
      "postDate": "01/21/2019 08:45:35",
      "content": "<p>Everyone is afraid in that sense, i am sure even the people on the top of LB must be</p>",
      "rawMarkdown": "Everyone is afraid in that sense, i am sure even the people on the top of LB must be",
      "votes": null
    },
    {
      "id": "459167",
      "postDate": "01/21/2019 10:08:59",
      "content": "<p>You're right. all of the people also afraid it haha :D. Thanks for advising me. </p>",
      "rawMarkdown": "You're right. all of the people also afraid it haha :D. Thanks for advising me.",
      "votes": null
    },
    {
      "id": "460217",
      "postDate": "01/23/2019 08:04:07",
      "content": "<p>Usually, if your score improves at each fold, you can get better score on LB. </p>",
      "rawMarkdown": "Usually, if your score improves at each fold, you can get better score on LB.",
      "votes": null
    },
    {
      "id": "460279",
      "postDate": "01/23/2019 10:59:45",
      "content": "<p>Thanks. your answer is best way in this situation. i keep in 1,2,3. (4 is memo...)\n- 1. Each fold : Better , LB : Better\n- 2. CV : Better, LB : Better\n- 3. CV : Better , LB : same\n- 4 CV : Not Better , Lb : Better</p>",
      "rawMarkdown": "Thanks. your answer is best way in this situation. i keep in 1,2,3. (4 is memo...)\n- 1. Each fold : Better , LB : Better\n- 2. CV : Better, LB : Better\n- 3. CV : Better , LB : same\n- 4 CV : Not Better , Lb : Better",
      "votes": null
    },
    {
      "id": "460348",
      "postDate": "01/23/2019 14:12:13",
      "content": "<p>hi ,can I ask you a quick question, do you keep track of scores  at each fold(such as: if you use 5 kfold,you note down the 5 scores) and compare them  after every training？</p>",
      "rawMarkdown": "hi ,can I ask you a quick question, do you keep track of scores  at each fold(such as: if you use 5 kfold,you note down the 5 scores) and compare them  after every training？",
      "votes": null
    },
    {
      "id": "460545",
      "postDate": "01/23/2019 23:42:11",
      "content": "<p>yes. check CV and each fold score. </p>",
      "rawMarkdown": "yes. check CV and each fold score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 458983,
      "author_name": "mks2192",
      "author_url": "",
      "post_date": "01/21/2019 00:17:52",
      "content": "<p>As a rational data scientist we tend to believe CV but it's really hard to say</p>\n\n<p><a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 458992,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/21/2019 01:03:22",
          "content": "<p>haha... thanks link and comment. i afraid that my model overfitting in the LB or Overfitting in the validation :(</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459129,
          "author_name": "mks2192",
          "author_url": "",
          "post_date": "01/21/2019 08:45:35",
          "content": "<p>Everyone is afraid in that sense, i am sure even the people on the top of LB must be</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459167,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/21/2019 10:08:59",
          "content": "<p>You're right. all of the people also afraid it haha :D. Thanks for advising me. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 460217,
      "author_name": "longyin2",
      "author_url": "",
      "post_date": "01/23/2019 08:04:07",
      "content": "<p>Usually, if your score improves at each fold, you can get better score on LB. </p>",
      "votes": null,
      "replies": [
        {
          "id": 460279,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/23/2019 10:59:45",
          "content": "<p>Thanks. your answer is best way in this situation. i keep in 1,2,3. (4 is memo...)\n- 1. Each fold : Better , LB : Better\n- 2. CV : Better, LB : Better\n- 3. CV : Better , LB : same\n- 4 CV : Not Better , Lb : Better</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460348,
          "author_name": "fangwenjun",
          "author_url": "",
          "post_date": "01/23/2019 14:12:13",
          "content": "<p>hi ,can I ask you a quick question, do you keep track of scores  at each fold(such as: if you use 5 kfold,you note down the 5 scores) and compare them  after every training？</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460545,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/23/2019 23:42:11",
          "content": "<p>yes. check CV and each fold score. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "458979": "9-fold LGB only - aggregation lr = 0.015\n\nI try to feature engineering and check to score. \nVersion1 : CV 3.6507 LB 3.688\nVersion2 :CV 3.6506 LB 3.689\nVersion3 :CV 3.6498 LB 3.689\nVersion4 :CV 3.6481 LB 3.690\nVersion5 :CV 3.6486 LB 3.693\n\nEven if my `CV is improved`, `LB is bad`. how should i do it..?",
    "458983": "As a rational data scientist we tend to believe CV but it's really hard to say\n\nhttps://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/77295",
    "458992": "haha... thanks link and comment. i afraid that my model overfitting in the LB or Overfitting in the validation :(",
    "459129": "Everyone is afraid in that sense, i am sure even the people on the top of LB must be",
    "459167": "You're right. all of the people also afraid it haha :D. Thanks for advising me.",
    "460217": "Usually, if your score improves at each fold, you can get better score on LB.",
    "460279": "Thanks. your answer is best way in this situation. i keep in 1,2,3. (4 is memo...)\n- 1. Each fold : Better , LB : Better\n- 2. CV : Better, LB : Better\n- 3. CV : Better , LB : same\n- 4 CV : Not Better , Lb : Better",
    "460348": "hi ,can I ask you a quick question, do you keep track of scores  at each fold(such as: if you use 5 kfold,you note down the 5 scores) and compare them  after every training？",
    "460545": "yes. check CV and each fold score."
  },
  "source": "meta"
}