{
  "id": 120376,
  "title": "actually i don't know  how i start this ?",
  "url": "/competitions/ashrae-energy-prediction/discussion/120376",
  "author_name": "",
  "post_date": "2019-12-05T15:45:32.696873400Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "",
  "messages": [
    {
      "id": "688448",
      "postDate": "12/05/2019 15:45:32",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "688520",
      "postDate": "12/05/2019 17:15:47",
      "content": "<p>Fairway:\nJust try to plot some histograms and scatter plots and see what each column means. Then submit a baseline, even if it is just the train data. Now think of features like interactions between weather, or even try to get some more data online like humidity or something. You could also see other notebooks and discussions to get some more ideas. Often I just end up multiplying various features and feeding it into the model as part of feature engineering. Towards the end of the competition, try out multiple models and ensemble them any way you find best.  You should probably be able to find the right set of params in notebooks anyway.</p>\n\n<p>Unfair way, but it is already out there:\nPeople have posted notebooks on leaked data, so you could follow those and get some. The discussions are also generally updated if someone finds another leak. So all you need to do is pretty much stay updated!</p>",
      "rawMarkdown": "Fairway:\nJust try to plot some histograms and scatter plots and see what each column means. Then submit a baseline, even if it is just the train data. Now think of features like interactions between weather, or even try to get some more data online like humidity or something. You could also see other notebooks and discussions to get some more ideas. Often I just end up multiplying various features and feeding it into the model as part of feature engineering. Towards the end of the competition, try out multiple models and ensemble them any way you find best.  You should probably be able to find the right set of params in notebooks anyway.\n\nUnfair way, but it is already out there:\nPeople have posted notebooks on leaked data, so you could follow those and get some. The discussions are also generally updated if someone finds another leak. So all you need to do is pretty much stay updated!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 688520,
      "author_name": "yushg123",
      "author_url": "",
      "post_date": "12/05/2019 17:15:47",
      "content": "<p>Fairway:\nJust try to plot some histograms and scatter plots and see what each column means. Then submit a baseline, even if it is just the train data. Now think of features like interactions between weather, or even try to get some more data online like humidity or something. You could also see other notebooks and discussions to get some more ideas. Often I just end up multiplying various features and feeding it into the model as part of feature engineering. Towards the end of the competition, try out multiple models and ensemble them any way you find best.  You should probably be able to find the right set of params in notebooks anyway.</p>\n\n<p>Unfair way, but it is already out there:\nPeople have posted notebooks on leaked data, so you could follow those and get some. The discussions are also generally updated if someone finds another leak. So all you need to do is pretty much stay updated!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "688448": "",
    "688520": "Fairway:\nJust try to plot some histograms and scatter plots and see what each column means. Then submit a baseline, even if it is just the train data. Now think of features like interactions between weather, or even try to get some more data online like humidity or something. You could also see other notebooks and discussions to get some more ideas. Often I just end up multiplying various features and feeding it into the model as part of feature engineering. Towards the end of the competition, try out multiple models and ensemble them any way you find best.  You should probably be able to find the right set of params in notebooks anyway.\n\nUnfair way, but it is already out there:\nPeople have posted notebooks on leaked data, so you could follow those and get some. The discussions are also generally updated if someone finds another leak. So all you need to do is pretty much stay updated!"
  },
  "source": "meta"
}