{
  "id": 77652,
  "title": "Feature engineering , CV/LB gap",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/77652",
  "author_name": "",
  "post_date": "2019-01-15T08:55:26.909756900Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have doing only feature engineering from Day_1. Till now , my single model LGB details are as below:\nCV- 3.6720684251809725  LB-3.684</p>\n\n<p>More features are reducing the CV but LB is constant at 3.84. None of the newly created feature is able to reduce the LB.</p>\n\n<p>All feature engineering are normal time delay , aggregation based on category and like that. I am sure that still there are scope for feature engineering.</p>\n\n<p>Feature elimination is not working. Libffm is not applied yet.\n\\\nAnything else to apply?</p>",
  "messages": [
    {
      "id": "456161",
      "postDate": "01/15/2019 08:55:26",
      "content": "<p>I have doing only feature engineering from Day_1. Till now , my single model LGB details are as below:\nCV- 3.6720684251809725  LB-3.684</p>\n\n<p>More features are reducing the CV but LB is constant at 3.84. None of the newly created feature is able to reduce the LB.</p>\n\n<p>All feature engineering are normal time delay , aggregation based on category and like that. I am sure that still there are scope for feature engineering.</p>\n\n<p>Feature elimination is not working. Libffm is not applied yet.\n\\\nAnything else to apply?</p>",
      "rawMarkdown": "I have doing only feature engineering from Day_1. Till now , my single model LGB details are as below:\nCV- 3.6720684251809725  LB-3.684\n\nMore features are reducing the CV but LB is constant at 3.84. None of the newly created feature is able to reduce the LB.\n\nAll feature engineering are normal time delay , aggregation based on category and like that. I am sure that still there are scope for feature engineering.\n\nFeature elimination is not working. Libffm is not applied yet.\n\\\nAnything else to apply?",
      "votes": null
    },
    {
      "id": "456163",
      "postDate": "01/15/2019 08:57:45",
      "content": "<p>Have you try on merchant.csv?</p>",
      "rawMarkdown": "Have you try on merchant.csv?",
      "votes": null
    },
    {
      "id": "456198",
      "postDate": "01/15/2019 10:32:32",
      "content": "<p>No..Nothing from merchant yet? are u successful to fetch some features from there?</p>",
      "rawMarkdown": "No..Nothing from merchant yet? are u successful to fetch some features from there?",
      "votes": null
    },
    {
      "id": "456971",
      "postDate": "01/16/2019 20:14:17",
      "content": "<p>I wonder for this competition, what does CV\\LB mean?\nmy CV folds have significant differences between them.\nSo I wonder if your CV 3.672 LB 3.684 is simply overfit of public LB.\nI am at about CV 3.645 LB 3.695~ with absolute consistency when CV go down, LB go down the same way.</p>\n\n<p>I wonder if shakeup in the end will replace our LB positions or not.\n=O</p>\n\n<p>Very tricky stuff</p>",
      "rawMarkdown": "I wonder for this competition, what does CV\\LB mean?\nmy CV folds have significant differences between them.\nSo I wonder if your CV 3.672 LB 3.684 is simply overfit of public LB.\nI am at about CV 3.645 LB 3.695~ with absolute consistency when CV go down, LB go down the same way.\n\nI wonder if shakeup in the end will replace our LB positions or not.\n=O\n\nVery tricky stuff",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 456163,
      "author_name": "marcuslin",
      "author_url": "",
      "post_date": "01/15/2019 08:57:45",
      "content": "<p>Have you try on merchant.csv?</p>",
      "votes": null,
      "replies": [
        {
          "id": 456198,
          "author_name": "subikashpal",
          "author_url": "",
          "post_date": "01/15/2019 10:32:32",
          "content": "<p>No..Nothing from merchant yet? are u successful to fetch some features from there?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 456971,
      "author_name": "tpthegreat",
      "author_url": "",
      "post_date": "01/16/2019 20:14:17",
      "content": "<p>I wonder for this competition, what does CV\\LB mean?\nmy CV folds have significant differences between them.\nSo I wonder if your CV 3.672 LB 3.684 is simply overfit of public LB.\nI am at about CV 3.645 LB 3.695~ with absolute consistency when CV go down, LB go down the same way.</p>\n\n<p>I wonder if shakeup in the end will replace our LB positions or not.\n=O</p>\n\n<p>Very tricky stuff</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "456161": "I have doing only feature engineering from Day_1. Till now , my single model LGB details are as below:\nCV- 3.6720684251809725  LB-3.684\n\nMore features are reducing the CV but LB is constant at 3.84. None of the newly created feature is able to reduce the LB.\n\nAll feature engineering are normal time delay , aggregation based on category and like that. I am sure that still there are scope for feature engineering.\n\nFeature elimination is not working. Libffm is not applied yet.\n\\\nAnything else to apply?",
    "456163": "Have you try on merchant.csv?",
    "456198": "No..Nothing from merchant yet? are u successful to fetch some features from there?",
    "456971": "I wonder for this competition, what does CV\\LB mean?\nmy CV folds have significant differences between them.\nSo I wonder if your CV 3.672 LB 3.684 is simply overfit of public LB.\nI am at about CV 3.645 LB 3.695~ with absolute consistency when CV go down, LB go down the same way.\n\nI wonder if shakeup in the end will replace our LB positions or not.\n=O\n\nVery tricky stuff"
  },
  "source": "meta"
}