{
  "id": 79775,
  "title": "Without removing or replacing outliers I got 1.55X CV but 3.84~3.86 LB",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/79775",
  "author_name": "Jie Wu",
  "post_date": "2019-02-07T12:09:37.605000",
  "votes": 0,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I created 360 features and use LGB to train the model including all outliers without any manipulation (on the target variable) I got 1.55X CV but worse LB betwen 3.84~3.86. </p>",
  "messages": [
    {
      "id": 469927,
      "postDate": "2019-02-12T04:19:07.260Z",
      "content": "<p>典型标签泄漏，做一些target相关的特征编码</p>",
      "rawMarkdown": "典型标签泄漏，做一些target相关的特征编码",
      "replies": [
        {
          "id": 470098,
          "postDate": "2019-02-12T11:18:00.200Z",
          "content": "<p>Yes, I found it. </p>",
          "rawMarkdown": "Yes, I found it. "
        }
      ]
    },
    {
      "id": 467613,
      "postDate": "2019-02-07T12:23:11.523Z",
      "content": "<p>Maybe you have information about outliers in your training set?</p>",
      "rawMarkdown": "Maybe you have information about outliers in your training set?",
      "replies": [
        {
          "id": 467616,
          "postDate": "2019-02-07T12:25:24.133Z",
          "content": "<p>Maybe. It worth to have a correlation analysis between target variable and my input variables.</p>",
          "rawMarkdown": "Maybe. It worth to have a correlation analysis between target variable and my input variables."
        },
        {
          "id": 467817,
          "postDate": "2019-02-07T19:26:44.290Z",
          "content": "<p>1.55x CV seems like CV for set without outliers\nmy actual CV 1.54x\nI think you should look on your sets (training and validation) - &gt; probably there are no outliers.\nalso lb score for full \"not outliers\" predictions is 3.870-3.877 (in my case)</p>",
          "rawMarkdown": "1.55x CV seems like CV for set without outliers\nmy actual CV 1.54x\nI think you should look on your sets (training and validation) - &gt; probably there are no outliers.\nalso lb score for full \"not outliers\" predictions is 3.870-3.877 (in my case)"
        },
        {
          "id": 467839,
          "postDate": "2019-02-07T20:27:30.437Z",
          "content": "<p>Yes this very strange, but I am very sure my model contains outliers</p>",
          "rawMarkdown": "Yes this very strange, but I am very sure my model contains outliers"
        }
      ]
    },
    {
      "id": 467605,
      "postDate": "2019-02-07T12:09:37.607Z",
      "content": "<p>I created 360 features and use LGB to train the model including all outliers without any manipulation (on the target variable) I got 1.55X CV but worse LB betwen 3.84~3.86. </p>",
      "rawMarkdown": "I created 360 features and use LGB to train the model including all outliers without any manipulation (on the target variable) I got 1.55X CV but worse LB betwen 3.84~3.86. "
    },
    {
      "id": 469545,
      "postDate": "2019-02-11T11:58:29.743Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 469927,
      "author_name": "Orange",
      "author_url": "",
      "post_date": "2019-02-12T04:19:07.260000",
      "content": "<p>典型标签泄漏，做一些target相关的特征编码</p>",
      "votes": 0,
      "replies": [
        {
          "id": 470098,
          "author_name": "Jie Wu",
          "author_url": "",
          "post_date": "2019-02-12T11:18:00.200000",
          "content": "<p>Yes, I found it. </p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 467613,
      "author_name": "Salamat",
      "author_url": "",
      "post_date": "2019-02-07T12:23:11.523000",
      "content": "<p>Maybe you have information about outliers in your training set?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 467616,
          "author_name": "Jie Wu",
          "author_url": "",
          "post_date": "2019-02-07T12:25:24.133000",
          "content": "<p>Maybe. It worth to have a correlation analysis between target variable and my input variables.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 467817,
          "author_name": "Konstantin Yakovlev",
          "author_url": "",
          "post_date": "2019-02-07T19:26:44.290000",
          "content": "<p>1.55x CV seems like CV for set without outliers\nmy actual CV 1.54x\nI think you should look on your sets (training and validation) - &gt; probably there are no outliers.\nalso lb score for full \"not outliers\" predictions is 3.870-3.877 (in my case)</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 467839,
          "author_name": "Jie Wu",
          "author_url": "",
          "post_date": "2019-02-07T20:27:30.437000",
          "content": "<p>Yes this very strange, but I am very sure my model contains outliers</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 469545,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-02-11T11:58:29.743000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "469927": "典型标签泄漏，做一些target相关的特征编码",
    "467613": "Maybe you have information about outliers in your training set?",
    "467605": "I created 360 features and use LGB to train the model including all outliers without any manipulation (on the target variable) I got 1.55X CV but worse LB betwen 3.84~3.86. ",
    "469545": ""
  }
}