{
  "id": 116446,
  "title": "Anyone had any luck feature engineering?",
  "url": "/competitions/ashrae-energy-prediction/discussion/116446",
  "author_name": "",
  "post_date": "2019-11-09T06:31:43.854363800Z",
  "votes": 2,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I know that I personally have gotten much more out of dropping observations and setting up weird configurations of models than feature engineering. I spent a whole day basically trying to find good features and my cv and lb only increased very very slightly.</p>\n\n<p>Has anybody had any luck with FE?</p>",
  "messages": [
    {
      "id": "668945",
      "postDate": "11/09/2019 06:31:43",
      "content": "<p>I know that I personally have gotten much more out of dropping observations and setting up weird configurations of models than feature engineering. I spent a whole day basically trying to find good features and my cv and lb only increased very very slightly.</p>\n\n<p>Has anybody had any luck with FE?</p>",
      "rawMarkdown": "I know that I personally have gotten much more out of dropping observations and setting up weird configurations of models than feature engineering. I spent a whole day basically trying to find good features and my cv and lb only increased very very slightly.\n\nHas anybody had any luck with FE?",
      "votes": null
    },
    {
      "id": "669082",
      "postDate": "11/09/2019 13:05:29",
      "content": "<p>Not really - importance of some of the lag features is non-zero, but it's pretty much in the tails of the distribution.</p>",
      "rawMarkdown": "Not really - importance of some of the lag features is non-zero, but it's pretty much in the tails of the distribution.",
      "votes": null
    },
    {
      "id": "669323",
      "postDate": "11/09/2019 21:42:25",
      "content": "<p>I’am still working on it... Be patient ;-)</p>",
      "rawMarkdown": "I’am still working on it... Be patient ;-)",
      "votes": null
    },
    {
      "id": "669336",
      "postDate": "11/09/2019 22:14:50",
      "content": "<p>Nope. I'm toying around with the 50/50 notebook and I've only gone from val's rmse : 1.5029 to around val's rmse : 1.49779 and I don't even know if it is reflected on the LB.</p>",
      "rawMarkdown": "Nope. I'm toying around with the 50/50 notebook and I've only gone from val's rmse : 1.5029 to around val's rmse : 1.49779 and I don't even know if it is reflected on the LB.",
      "votes": null
    },
    {
      "id": "669353",
      "postDate": "11/09/2019 23:21:05",
      "content": "<p>Funny, 1.49779 is more or less what I got too at the best ;-)</p>\n\n<p>But I suspect this is not the <strong>real truth</strong></p>",
      "rawMarkdown": "Funny, 1.49779 is more or less what I got too at the best ;-)\n\nBut I suspect this is not the **real truth**",
      "votes": null
    },
    {
      "id": "669964",
      "postDate": "11/10/2019 19:23:27",
      "content": "<p>Very true. Models fail to predict correctly very high and very low values.\n<a href=\"https://www.kaggle.com/hmendonca/4-ashrae-blended\">https://www.kaggle.com/hmendonca/4-ashrae-blended</a></p>",
      "rawMarkdown": "Very true. Models fail to predict correctly very high and very low values.\nhttps://www.kaggle.com/hmendonca/4-ashrae-blended",
      "votes": null
    },
    {
      "id": "670138",
      "postDate": "11/11/2019 04:14:32",
      "content": "<p>hmmm very cryptic... what do you mean bv the \"real truth\"?</p>",
      "rawMarkdown": "hmmm very cryptic... what do you mean bv the \"real truth\"?",
      "votes": null
    },
    {
      "id": "670253",
      "postDate": "11/11/2019 08:14:43",
      "content": "<p>I removed the strange site_id 0 data and validation went from  1.5029 to 1.144, but LB went to 1.43...</p>",
      "rawMarkdown": "I removed the strange site_id 0 data and validation went from  1.5029 to 1.144, but LB went to 1.43...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 669082,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "11/09/2019 13:05:29",
      "content": "<p>Not really - importance of some of the lag features is non-zero, but it's pretty much in the tails of the distribution.</p>",
      "votes": null,
      "replies": [
        {
          "id": 669964,
          "author_name": "vishal1310",
          "author_url": "",
          "post_date": "11/10/2019 19:23:27",
          "content": "<p>Very true. Models fail to predict correctly very high and very low values.\n<a href=\"https://www.kaggle.com/hmendonca/4-ashrae-blended\">https://www.kaggle.com/hmendonca/4-ashrae-blended</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 669323,
      "author_name": "pfichou",
      "author_url": "",
      "post_date": "11/09/2019 21:42:25",
      "content": "<p>I’am still working on it... Be patient ;-)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 669336,
      "author_name": "frjo0083",
      "author_url": "",
      "post_date": "11/09/2019 22:14:50",
      "content": "<p>Nope. I'm toying around with the 50/50 notebook and I've only gone from val's rmse : 1.5029 to around val's rmse : 1.49779 and I don't even know if it is reflected on the LB.</p>",
      "votes": null,
      "replies": [
        {
          "id": 669353,
          "author_name": "pfichou",
          "author_url": "",
          "post_date": "11/09/2019 23:21:05",
          "content": "<p>Funny, 1.49779 is more or less what I got too at the best ;-)</p>\n\n<p>But I suspect this is not the <strong>real truth</strong></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 670138,
          "author_name": "lewington",
          "author_url": "",
          "post_date": "11/11/2019 04:14:32",
          "content": "<p>hmmm very cryptic... what do you mean bv the \"real truth\"?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 670253,
          "author_name": "frjo0083",
          "author_url": "",
          "post_date": "11/11/2019 08:14:43",
          "content": "<p>I removed the strange site_id 0 data and validation went from  1.5029 to 1.144, but LB went to 1.43...</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "668945": "I know that I personally have gotten much more out of dropping observations and setting up weird configurations of models than feature engineering. I spent a whole day basically trying to find good features and my cv and lb only increased very very slightly.\n\nHas anybody had any luck with FE?",
    "669082": "Not really - importance of some of the lag features is non-zero, but it's pretty much in the tails of the distribution.",
    "669323": "I’am still working on it... Be patient ;-)",
    "669336": "Nope. I'm toying around with the 50/50 notebook and I've only gone from val's rmse : 1.5029 to around val's rmse : 1.49779 and I don't even know if it is reflected on the LB.",
    "669353": "Funny, 1.49779 is more or less what I got too at the best ;-)\n\nBut I suspect this is not the **real truth**",
    "669964": "Very true. Models fail to predict correctly very high and very low values.\nhttps://www.kaggle.com/hmendonca/4-ashrae-blended",
    "670138": "hmmm very cryptic... what do you mean bv the \"real truth\"?",
    "670253": "I removed the strange site_id 0 data and validation went from  1.5029 to 1.144, but LB went to 1.43..."
  },
  "source": "meta"
}