{
  "id": 115022,
  "title": "I would venture that private LB is actually Q1'2018",
  "url": "/competitions/ashrae-energy-prediction/discussion/115022",
  "author_name": "",
  "post_date": "2019-10-30T19:12:47.693033500Z",
  "votes": 6,
  "comment_count": 8,
  "views": 0,
  "content": "<p>And the rest is there just for fun\nThat would fit the 22% narrative, so that in total the test set is only 2017-2018Q1</p>",
  "messages": [
    {
      "id": "661852",
      "postDate": "10/30/2019 19:12:47",
      "content": "<p>And the rest is there just for fun\nThat would fit the 22% narrative, so that in total the test set is only 2017-2018Q1</p>",
      "rawMarkdown": "And the rest is there just for fun\nThat would fit the 22% narrative, so that in total the test set is only 2017-2018Q1",
      "votes": null
    },
    {
      "id": "662093",
      "postDate": "10/31/2019 04:29:42",
      "content": "<p>That's hard to predict. Its better to stay prepared for anything unexpected. Who knows how much it would shake up in private LB 😅 </p>",
      "rawMarkdown": "That's hard to predict. Its better to stay prepared for anything unexpected. Who knows how much it would shake up in private LB 😅",
      "votes": null
    },
    {
      "id": "662255",
      "postDate": "10/31/2019 10:20:17",
      "content": "<p>Very interesting indeed, we can do a comparison with the external open source data.</p>",
      "rawMarkdown": "Very interesting indeed, we can do a comparison with the external open source data.",
      "votes": null
    },
    {
      "id": "662384",
      "postDate": "10/31/2019 13:48:08",
      "content": "<p>If this is true, it might help explain why they only gave us 1 year of training data instead of 2, which seems like the obvious thing to do if there's actually 3 full years of data. I hope it's not true though, if only for the reason that forcing us to generate all those extra, unused predictions for 2018 would be so inefficient.</p>",
      "rawMarkdown": "If this is true, it might help explain why they only gave us 1 year of training data instead of 2, which seems like the obvious thing to do if there's actually 3 full years of data. I hope it's not true though, if only for the reason that forcing us to generate all those extra, unused predictions for 2018 would be so inefficient.",
      "votes": null
    },
    {
      "id": "662414",
      "postDate": "10/31/2019 14:29:18",
      "content": "<p>So you think that 2017 is the public set, Q1 2018 is the private set, and Q2-3-4 2018 is not used? </p>\n\n<p>To me that sounds unlikely, because we would be skipping summer, where most of the cooling action occurs!</p>",
      "rawMarkdown": "So you think that 2017 is the public set, Q1 2018 is the private set, and Q2-3-4 2018 is not used? \n\nTo me that sounds unlikely, because we would be skipping summer, where most of the cooling action occurs!",
      "votes": null
    },
    {
      "id": "665258",
      "postDate": "11/04/2019 20:04:35",
      "content": "<p>Reading the notes from the organizers:</p>\n\n<p>We represent a small, diverse and dedicated group of volunteers who have been working together over the past 16 months to create the relationships and pull together the necessary resources to launch a successful Kaggle data competition.</p>\n\n<p>If it took them 16 months to put together this competition, then the dataset can in fact end somewhere around 2018Q2 and that might support the point of 2018Q1 being the private LB</p>",
      "rawMarkdown": "Reading the notes from the organizers:\n\nWe represent a small, diverse and dedicated group of volunteers who have been working together over the past 16 months to create the relationships and pull together the necessary resources to launch a successful Kaggle data competition.\n\nIf it took them 16 months to put together this competition, then the dataset can in fact end somewhere around 2018Q2 and that might support the point of 2018Q1 being the private LB",
      "votes": null
    },
    {
      "id": "665766",
      "postDate": "11/05/2019 11:32:50",
      "content": "<p>A simple way to test this assumption would be to set values for submitted predictions outside to 18Q1 to zero and see if the LB score changes. Has anyone tried that?</p>",
      "rawMarkdown": "A simple way to test this assumption would be to set values for submitted predictions outside to 18Q1 to zero and see if the LB score changes. Has anyone tried that?",
      "votes": null
    },
    {
      "id": "665774",
      "postDate": "11/05/2019 11:45:50",
      "content": "<p>I enjoyed this keen sense of a detective :)</p>",
      "rawMarkdown": "I enjoyed this keen sense of a detective :)",
      "votes": null
    },
    {
      "id": "666394",
      "postDate": "11/06/2019 04:07:59",
      "content": "<p>On private LB? We need to wait 5 weeks still</p>",
      "rawMarkdown": "On private LB? We need to wait 5 weeks still",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 662093,
      "author_name": "khairulislam",
      "author_url": "",
      "post_date": "10/31/2019 04:29:42",
      "content": "<p>That's hard to predict. Its better to stay prepared for anything unexpected. Who knows how much it would shake up in private LB 😅 </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 662255,
      "author_name": "s13658695",
      "author_url": "",
      "post_date": "10/31/2019 10:20:17",
      "content": "<p>Very interesting indeed, we can do a comparison with the external open source data.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 662384,
      "author_name": "aquatic",
      "author_url": "",
      "post_date": "10/31/2019 13:48:08",
      "content": "<p>If this is true, it might help explain why they only gave us 1 year of training data instead of 2, which seems like the obvious thing to do if there's actually 3 full years of data. I hope it's not true though, if only for the reason that forcing us to generate all those extra, unused predictions for 2018 would be so inefficient.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 662414,
      "author_name": "sdoria",
      "author_url": "",
      "post_date": "10/31/2019 14:29:18",
      "content": "<p>So you think that 2017 is the public set, Q1 2018 is the private set, and Q2-3-4 2018 is not used? </p>\n\n<p>To me that sounds unlikely, because we would be skipping summer, where most of the cooling action occurs!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 665258,
      "author_name": "snovik1975",
      "author_url": "",
      "post_date": "11/04/2019 20:04:35",
      "content": "<p>Reading the notes from the organizers:</p>\n\n<p>We represent a small, diverse and dedicated group of volunteers who have been working together over the past 16 months to create the relationships and pull together the necessary resources to launch a successful Kaggle data competition.</p>\n\n<p>If it took them 16 months to put together this competition, then the dataset can in fact end somewhere around 2018Q2 and that might support the point of 2018Q1 being the private LB</p>",
      "votes": null,
      "replies": [
        {
          "id": 665774,
          "author_name": "barnwellguy",
          "author_url": "",
          "post_date": "11/05/2019 11:45:50",
          "content": "<p>I enjoyed this keen sense of a detective :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 665766,
      "author_name": "poedator",
      "author_url": "",
      "post_date": "11/05/2019 11:32:50",
      "content": "<p>A simple way to test this assumption would be to set values for submitted predictions outside to 18Q1 to zero and see if the LB score changes. Has anyone tried that?</p>",
      "votes": null,
      "replies": [
        {
          "id": 666394,
          "author_name": "snovik1975",
          "author_url": "",
          "post_date": "11/06/2019 04:07:59",
          "content": "<p>On private LB? We need to wait 5 weeks still</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "661852": "And the rest is there just for fun\nThat would fit the 22% narrative, so that in total the test set is only 2017-2018Q1",
    "662093": "That's hard to predict. Its better to stay prepared for anything unexpected. Who knows how much it would shake up in private LB 😅",
    "662255": "Very interesting indeed, we can do a comparison with the external open source data.",
    "662384": "If this is true, it might help explain why they only gave us 1 year of training data instead of 2, which seems like the obvious thing to do if there's actually 3 full years of data. I hope it's not true though, if only for the reason that forcing us to generate all those extra, unused predictions for 2018 would be so inefficient.",
    "662414": "So you think that 2017 is the public set, Q1 2018 is the private set, and Q2-3-4 2018 is not used? \n\nTo me that sounds unlikely, because we would be skipping summer, where most of the cooling action occurs!",
    "665258": "Reading the notes from the organizers:\n\nWe represent a small, diverse and dedicated group of volunteers who have been working together over the past 16 months to create the relationships and pull together the necessary resources to launch a successful Kaggle data competition.\n\nIf it took them 16 months to put together this competition, then the dataset can in fact end somewhere around 2018Q2 and that might support the point of 2018Q1 being the private LB",
    "665766": "A simple way to test this assumption would be to set values for submitted predictions outside to 18Q1 to zero and see if the LB score changes. Has anyone tried that?",
    "665774": "I enjoyed this keen sense of a detective :)",
    "666394": "On private LB? We need to wait 5 weeks still"
  },
  "source": "meta"
}