{
  "id": 114581,
  "title": "A few Interesting Findings",
  "url": "/competitions/ashrae-energy-prediction/discussion/114581",
  "author_name": "",
  "post_date": "2019-10-27T17:38:42.245697800Z",
  "votes": 9,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I was doing some bit of basic EDA on my own and I found a few interesting things:\nThe meter reading of a few houses has not been recorded for some days or hours of the year. Won't this affect our model, and if so what can we do about it?</p>\n\n<p>Moreover, I tried submitting the target variable of the train data and got a public lb score of 1.78. I did this by matching the day and month of the year from the train set to the test set.\nDoesn't this show that there is not much change in energy usage over the years, for the same house?</p>\n\n<p>Thanks</p>",
  "messages": [
    {
      "id": "659453",
      "postDate": "10/27/2019 17:38:42",
      "content": "<p>I was doing some bit of basic EDA on my own and I found a few interesting things:\nThe meter reading of a few houses has not been recorded for some days or hours of the year. Won't this affect our model, and if so what can we do about it?</p>\n\n<p>Moreover, I tried submitting the target variable of the train data and got a public lb score of 1.78. I did this by matching the day and month of the year from the train set to the test set.\nDoesn't this show that there is not much change in energy usage over the years, for the same house?</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "I was doing some bit of basic EDA on my own and I found a few interesting things:\nThe meter reading of a few houses has not been recorded for some days or hours of the year. Won't this affect our model, and if so what can we do about it?\n\nMoreover, I tried submitting the target variable of the train data and got a public lb score of 1.78. I did this by matching the day and month of the year from the train set to the test set.\nDoesn't this show that there is not much change in energy usage over the years, for the same house?\n\nThanks",
      "votes": null
    },
    {
      "id": "659592",
      "postDate": "10/28/2019 00:36:47",
      "content": "<p>I'm doing an <a href=\"https://www.kaggle.com/juanmah/ashrae-gaps\">EDA notebook</a> to study the gaps in the train set. Currently, it is WIP. I want to get some conclusions in the next days.</p>\n\n<p>The gaps WILL affect the model, because the lack of data. The lacking data SHOULD be interpolated to minimize the error of the model with the reality. Quantifying the affectation is a point to consider.</p>\n\n<p>The buildings have, normally, a daily pattern (if the day is a working day, weekend or a bank holiday), a weekly pattern (every day has some nuances, for example, Fridays), and a monthly pattern (cold months has a different pattern than warm months). If you match the day_of_the_year from train to set it could be a good start point. If you align to match also the day of the week, you will have a better model.</p>",
      "rawMarkdown": "I'm doing an [EDA notebook](https://www.kaggle.com/juanmah/ashrae-gaps) to study the gaps in the train set. Currently, it is WIP. I want to get some conclusions in the next days.\n\nThe gaps WILL affect the model, because the lack of data. The lacking data SHOULD be interpolated to minimize the error of the model with the reality. Quantifying the affectation is a point to consider.\n\nThe buildings have, normally, a daily pattern (if the day is a working day, weekend or a bank holiday), a weekly pattern (every day has some nuances, for example, Fridays), and a monthly pattern (cold months has a different pattern than warm months). If you match the day_of_the_year from train to set it could be a good start point. If you align to match also the day of the week, you will have a better model.",
      "votes": null
    },
    {
      "id": "659912",
      "postDate": "10/28/2019 12:34:44",
      "content": "<p>Thanks a lot. I now have some direction on what to do with the gaps.</p>",
      "rawMarkdown": "Thanks a lot. I now have some direction on what to do with the gaps.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 659592,
      "author_name": "juanmah",
      "author_url": "",
      "post_date": "10/28/2019 00:36:47",
      "content": "<p>I'm doing an <a href=\"https://www.kaggle.com/juanmah/ashrae-gaps\">EDA notebook</a> to study the gaps in the train set. Currently, it is WIP. I want to get some conclusions in the next days.</p>\n\n<p>The gaps WILL affect the model, because the lack of data. The lacking data SHOULD be interpolated to minimize the error of the model with the reality. Quantifying the affectation is a point to consider.</p>\n\n<p>The buildings have, normally, a daily pattern (if the day is a working day, weekend or a bank holiday), a weekly pattern (every day has some nuances, for example, Fridays), and a monthly pattern (cold months has a different pattern than warm months). If you match the day_of_the_year from train to set it could be a good start point. If you align to match also the day of the week, you will have a better model.</p>",
      "votes": null,
      "replies": [
        {
          "id": 659912,
          "author_name": "yushg123",
          "author_url": "",
          "post_date": "10/28/2019 12:34:44",
          "content": "<p>Thanks a lot. I now have some direction on what to do with the gaps.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "659453": "I was doing some bit of basic EDA on my own and I found a few interesting things:\nThe meter reading of a few houses has not been recorded for some days or hours of the year. Won't this affect our model, and if so what can we do about it?\n\nMoreover, I tried submitting the target variable of the train data and got a public lb score of 1.78. I did this by matching the day and month of the year from the train set to the test set.\nDoesn't this show that there is not much change in energy usage over the years, for the same house?\n\nThanks",
    "659592": "I'm doing an [EDA notebook](https://www.kaggle.com/juanmah/ashrae-gaps) to study the gaps in the train set. Currently, it is WIP. I want to get some conclusions in the next days.\n\nThe gaps WILL affect the model, because the lack of data. The lacking data SHOULD be interpolated to minimize the error of the model with the reality. Quantifying the affectation is a point to consider.\n\nThe buildings have, normally, a daily pattern (if the day is a working day, weekend or a bank holiday), a weekly pattern (every day has some nuances, for example, Fridays), and a monthly pattern (cold months has a different pattern than warm months). If you match the day_of_the_year from train to set it could be a good start point. If you align to match also the day of the week, you will have a better model.",
    "659912": "Thanks a lot. I now have some direction on what to do with the gaps."
  },
  "source": "meta"
}