{
  "id": 69919,
  "title": "Normalization of the Light Curves",
  "url": "/competitions/PLAsTiCC-2018/discussion/69919",
  "author_name": "",
  "post_date": "2018-10-28T20:47:27.161647Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>We are wondering what is the best way to normalize the different light curves across the different bands. We think different bands should be normalize together, since the ordering is important. However the standard deviation is quite big so we don't know if using the standard deviation of the whole training set across all bands is the way to go. Another option is normalize individually every example, or maybe use a logarithmic scale. Any thoughts?</p>",
  "messages": [
    {
      "id": "411707",
      "postDate": "10/28/2018 20:47:27",
      "content": "<p>We are wondering what is the best way to normalize the different light curves across the different bands. We think different bands should be normalize together, since the ordering is important. However the standard deviation is quite big so we don't know if using the standard deviation of the whole training set across all bands is the way to go. Another option is normalize individually every example, or maybe use a logarithmic scale. Any thoughts?</p>",
      "rawMarkdown": "We are wondering what is the best way to normalize the different light curves across the different bands. We think different bands should be normalize together, since the ordering is important. However the standard deviation is quite big so we don't know if using the standard deviation of the whole training set across all bands is the way to go. Another option is normalize individually every example, or maybe use a logarithmic scale. Any thoughts?",
      "votes": null
    },
    {
      "id": "411783",
      "postDate": "10/29/2018 02:38:32",
      "content": "<p>Try everything, and then choose the best.</p>\n\n<p>(I wouldn't say choosing one is the best way)</p>",
      "rawMarkdown": "Try everything, and then choose the best.\n\n(I wouldn't say choosing one is the best way)",
      "votes": null
    },
    {
      "id": "411925",
      "postDate": "10/29/2018 08:12:00",
      "content": "<p>Interpolation + linear regression for restoring curve shape\n<a href=\"https://www.kaggle.com/sergeylebedev/light-curve-equalization\">https://www.kaggle.com/sergeylebedev/light-curve-equalization</a></p>",
      "rawMarkdown": "Interpolation + linear regression for restoring curve shape\nhttps://www.kaggle.com/sergeylebedev/light-curve-equalization",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 411783,
      "author_name": "onodera",
      "author_url": "",
      "post_date": "10/29/2018 02:38:32",
      "content": "<p>Try everything, and then choose the best.</p>\n\n<p>(I wouldn't say choosing one is the best way)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 411925,
      "author_name": "sergeylebedev",
      "author_url": "",
      "post_date": "10/29/2018 08:12:00",
      "content": "<p>Interpolation + linear regression for restoring curve shape\n<a href=\"https://www.kaggle.com/sergeylebedev/light-curve-equalization\">https://www.kaggle.com/sergeylebedev/light-curve-equalization</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "411707": "We are wondering what is the best way to normalize the different light curves across the different bands. We think different bands should be normalize together, since the ordering is important. However the standard deviation is quite big so we don't know if using the standard deviation of the whole training set across all bands is the way to go. Another option is normalize individually every example, or maybe use a logarithmic scale. Any thoughts?",
    "411783": "Try everything, and then choose the best.\n\n(I wouldn't say choosing one is the best way)",
    "411925": "Interpolation + linear regression for restoring curve shape\nhttps://www.kaggle.com/sergeylebedev/light-curve-equalization"
  },
  "source": "meta"
}