{
  "id": 55523,
  "title": "Reason behind Notebook gives better result? LB: 0.9786",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/55523",
  "author_name": "",
  "post_date": "2018-04-27T21:20:11.636340100Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi, I have modified my Notebook version. I did same with normal Script version also. But with same code I got slightly better result with my notebook version than <a href=\"https://www.kaggle.com/asraful70/talkingdata-new-features-in-lightgbm-lb-0-9784\">Script</a> version. What could be the reason there? Or its just because of LightGBM's random behavior.  Check out my Notebook version on the following this link:\n<a href=\"https://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786\">https://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786</a></p>",
  "messages": [
    {
      "id": "320207",
      "postDate": "04/27/2018 21:20:11",
      "content": "<p>Hi, I have modified my Notebook version. I did same with normal Script version also. But with same code I got slightly better result with my notebook version than <a href=\"https://www.kaggle.com/asraful70/talkingdata-new-features-in-lightgbm-lb-0-9784\">Script</a> version. What could be the reason there? Or its just because of LightGBM's random behavior.  Check out my Notebook version on the following this link:\n<a href=\"https://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786\">https://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786</a></p>",
      "rawMarkdown": "Hi, I have modified my Notebook version. I did same with normal Script version also. But with same code I got slightly better result with my notebook version than [Script][1] version. What could be the reason there? Or its just because of LightGBM's random behavior.  Check out my Notebook version on the following this link:\nhttps://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786\n\n\n  [1]: https://www.kaggle.com/asraful70/talkingdata-new-features-in-lightgbm-lb-0-9784",
      "votes": null
    },
    {
      "id": "320226",
      "postDate": "04/27/2018 23:25:10",
      "content": "<p>Since the code is the same, it's just the Kaggle Random Gods gifting you a couple of points.</p>\n\n<p>Even though the increase was just random noise, you can use this to your advantage: usually when you run and average the results of the same script multiple times (with different seed), the noise tends to be minimized.</p>",
      "rawMarkdown": "Since the code is the same, it's just the Kaggle Random Gods gifting you a couple of points.\n\nEven though the increase was just random noise, you can use this to your advantage: usually when you run and average the results of the same script multiple times (with different seed), the noise tends to be minimized.",
      "votes": null
    },
    {
      "id": "320367",
      "postDate": "04/28/2018 13:00:46",
      "content": "<p>I also thought so. Thanks for your clarification.</p>",
      "rawMarkdown": "I also thought so. Thanks for your clarification.",
      "votes": null
    },
    {
      "id": "320444",
      "postDate": "04/28/2018 17:40:58",
      "content": "<p>Would it be possible that different (18%)  test samples are used for calculating the public score?</p>",
      "rawMarkdown": "Would it be possible that different (18%)  test samples are used for calculating the public score?",
      "votes": null
    },
    {
      "id": "320473",
      "postDate": "04/28/2018 19:08:03",
      "content": "<p>i do 't think so. its automatically selected</p>",
      "rawMarkdown": "i do 't think so. its automatically selected",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 320226,
      "author_name": "antmarakis",
      "author_url": "",
      "post_date": "04/27/2018 23:25:10",
      "content": "<p>Since the code is the same, it's just the Kaggle Random Gods gifting you a couple of points.</p>\n\n<p>Even though the increase was just random noise, you can use this to your advantage: usually when you run and average the results of the same script multiple times (with different seed), the noise tends to be minimized.</p>",
      "votes": null,
      "replies": [
        {
          "id": 320367,
          "author_name": "asraful70",
          "author_url": "",
          "post_date": "04/28/2018 13:00:46",
          "content": "<p>I also thought so. Thanks for your clarification.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 320444,
      "author_name": "tapioca",
      "author_url": "",
      "post_date": "04/28/2018 17:40:58",
      "content": "<p>Would it be possible that different (18%)  test samples are used for calculating the public score?</p>",
      "votes": null,
      "replies": [
        {
          "id": 320473,
          "author_name": "asraful70",
          "author_url": "",
          "post_date": "04/28/2018 19:08:03",
          "content": "<p>i do 't think so. its automatically selected</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "320207": "Hi, I have modified my Notebook version. I did same with normal Script version also. But with same code I got slightly better result with my notebook version than [Script][1] version. What could be the reason there? Or its just because of LightGBM's random behavior.  Check out my Notebook version on the following this link:\nhttps://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lightgbm-lb-0-9786\n\n\n  [1]: https://www.kaggle.com/asraful70/talkingdata-new-features-in-lightgbm-lb-0-9784",
    "320226": "Since the code is the same, it's just the Kaggle Random Gods gifting you a couple of points.\n\nEven though the increase was just random noise, you can use this to your advantage: usually when you run and average the results of the same script multiple times (with different seed), the noise tends to be minimized.",
    "320367": "I also thought so. Thanks for your clarification.",
    "320444": "Would it be possible that different (18%)  test samples are used for calculating the public score?",
    "320473": "i do 't think so. its automatically selected"
  },
  "source": "meta"
}