{
  "id": 117937,
  "title": "Has anyone tried using the leaked data to get more training years?",
  "url": "/competitions/ashrae-energy-prediction/discussion/117937",
  "author_name": "",
  "post_date": "2019-11-18T21:37:47.170488400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I'm wondering if anyone has used the leaked data to look at year-over-year patterns in the data. It's a little frustrating that we've only got one year to work with, as I'd expect this data to be pretty consistent over the years. I'm wondering if anyone has used the leaked data to get earlier years to train on.</p>",
  "messages": [
    {
      "id": "675991",
      "postDate": "11/18/2019 21:37:47",
      "content": "<p>I'm wondering if anyone has used the leaked data to look at year-over-year patterns in the data. It's a little frustrating that we've only got one year to work with, as I'd expect this data to be pretty consistent over the years. I'm wondering if anyone has used the leaked data to get earlier years to train on.</p>",
      "rawMarkdown": "I'm wondering if anyone has used the leaked data to look at year-over-year patterns in the data. It's a little frustrating that we've only got one year to work with, as I'd expect this data to be pretty consistent over the years. I'm wondering if anyone has used the leaked data to get earlier years to train on.",
      "votes": null
    },
    {
      "id": "675995",
      "postDate": "11/18/2019 21:50:01",
      "content": "<p>I did - I used to add a \"month\" feature, but so far it just overfits like crazy.</p>",
      "rawMarkdown": "I did - I used to add a \"month\" feature, but so far it just overfits like crazy.",
      "votes": null
    },
    {
      "id": "676128",
      "postDate": "11/19/2019 01:22:22",
      "content": "<p>I just used the public leakage data to fill the submission csv. \nbut I think the final score do not contain the leakage data. I try to use [Building gene... ]GitHub data 2015 ] to train. but it's don't help me to approve. for now , just forget the public score. to do what you want way to approve the score.  I'm pure kernel just get 1.06 in Public . so keep going you can get a good score.</p>",
      "rawMarkdown": "I just used the public leakage data to fill the submission csv. \nbut I think the final score do not contain the leakage data. I try to use [Building gene... ]GitHub data 2015 ] to train. but it's don't help me to approve. for now , just forget the public score. to do what you want way to approve the score.  I'm pure kernel just get 1.06 in Public . so keep going you can get a good score.",
      "votes": null
    },
    {
      "id": "677688",
      "postDate": "11/20/2019 13:45:16",
      "content": "<p>I also want to know</p>",
      "rawMarkdown": "I also want to know",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 675995,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "11/18/2019 21:50:01",
      "content": "<p>I did - I used to add a \"month\" feature, but so far it just overfits like crazy.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 676128,
      "author_name": "pp2file",
      "author_url": "",
      "post_date": "11/19/2019 01:22:22",
      "content": "<p>I just used the public leakage data to fill the submission csv. \nbut I think the final score do not contain the leakage data. I try to use [Building gene... ]GitHub data 2015 ] to train. but it's don't help me to approve. for now , just forget the public score. to do what you want way to approve the score.  I'm pure kernel just get 1.06 in Public . so keep going you can get a good score.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 677688,
      "author_name": "zidaoziyan",
      "author_url": "",
      "post_date": "11/20/2019 13:45:16",
      "content": "<p>I also want to know</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "675991": "I'm wondering if anyone has used the leaked data to look at year-over-year patterns in the data. It's a little frustrating that we've only got one year to work with, as I'd expect this data to be pretty consistent over the years. I'm wondering if anyone has used the leaked data to get earlier years to train on.",
    "675995": "I did - I used to add a \"month\" feature, but so far it just overfits like crazy.",
    "676128": "I just used the public leakage data to fill the submission csv. \nbut I think the final score do not contain the leakage data. I try to use [Building gene... ]GitHub data 2015 ] to train. but it's don't help me to approve. for now , just forget the public score. to do what you want way to approve the score.  I'm pure kernel just get 1.06 in Public . so keep going you can get a good score.",
    "677688": "I also want to know"
  },
  "source": "meta"
}