{
  "id": 76146,
  "title": "Fourier and periodogram features",
  "url": "/competitions/PLAsTiCC-2018/discussion/76146",
  "author_name": "",
  "post_date": "2018-12-29T18:20:50.381054700Z",
  "votes": 6,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Sadly, we joined in a bit late and were quite busy so we couldn't enjoy so much this very exciting competition. However, we think we did implement some things that could be interesting/useful for others. Firstly, we re-implemented paper [1] in python to extract Fourier space features from unevenly sampled data in time domain. The code can be found at [2] and it is relatively fast. It computes the magnitude, unwrapped phase, periodogram and true alarm probability of a given period to come from an actual signal. </p>\n\n<p><img src=\"https://github.com/edumotya/pyriodogram/blob/master/example.png?raw=true\" alt=\"Features for plasticc dataset with object_id 612\"></p>\n\n<p>To avoid recomputing something we have already done, we attach the hdf5 file containing Kyle's augmented dataset (thanks for sharing it by the way, it's great) together with the Fourier features [3].  This file can be read using the fold number as a key:</p>\n\n<p>df = pd.read_hdf('fourier_feats.hdf', 'fold_0')</p>\n\n<p>[1] <a href=\"https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf\">https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf</a></p>\n\n<p>[2] <a href=\"https://github.com/edumotya/pyriodogram\">https://github.com/edumotya/pyriodogram</a></p>\n\n<p>[3] <a href=\"https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing\">https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing</a></p>",
  "messages": [
    {
      "id": "447384",
      "postDate": "12/29/2018 18:20:50",
      "content": "<p>Sadly, we joined in a bit late and were quite busy so we couldn't enjoy so much this very exciting competition. However, we think we did implement some things that could be interesting/useful for others. Firstly, we re-implemented paper [1] in python to extract Fourier space features from unevenly sampled data in time domain. The code can be found at [2] and it is relatively fast. It computes the magnitude, unwrapped phase, periodogram and true alarm probability of a given period to come from an actual signal. </p>\n\n<p><img src=\"https://github.com/edumotya/pyriodogram/blob/master/example.png?raw=true\" alt=\"Features for plasticc dataset with object_id 612\"></p>\n\n<p>To avoid recomputing something we have already done, we attach the hdf5 file containing Kyle's augmented dataset (thanks for sharing it by the way, it's great) together with the Fourier features [3].  This file can be read using the fold number as a key:</p>\n\n<p>df = pd.read_hdf('fourier_feats.hdf', 'fold_0')</p>\n\n<p>[1] <a href=\"https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf\">https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf</a></p>\n\n<p>[2] <a href=\"https://github.com/edumotya/pyriodogram\">https://github.com/edumotya/pyriodogram</a></p>\n\n<p>[3] <a href=\"https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing\">https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing</a></p>",
      "rawMarkdown": "Sadly, we joined in a bit late and were quite busy so we couldn't enjoy so much this very exciting competition. However, we think we did implement some things that could be interesting/useful for others. Firstly, we re-implemented paper [1] in python to extract Fourier space features from unevenly sampled data in time domain. The code can be found at [2] and it is relatively fast. It computes the magnitude, unwrapped phase, periodogram and true alarm probability of a given period to come from an actual signal. \n\n![Features for plasticc dataset with object_id 612][4]\n\nTo avoid recomputing something we have already done, we attach the hdf5 file containing Kyle's augmented dataset (thanks for sharing it by the way, it's great) together with the Fourier features [3].  This file can be read using the fold number as a key:\n\ndf = pd.read_hdf('fourier_feats.hdf', 'fold_0')\n\n[1] https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf\n\n[2] https://github.com/edumotya/pyriodogram\n\n[3] https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing\n\n\n  [4]: https://github.com/edumotya/pyriodogram/blob/master/example.png?raw=true",
      "votes": null
    },
    {
      "id": "448116",
      "postDate": "12/31/2018 09:23:26",
      "content": "<p>Thanks for sharing.</p>",
      "rawMarkdown": "Thanks for sharing.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 448116,
      "author_name": "longyin2",
      "author_url": "",
      "post_date": "12/31/2018 09:23:26",
      "content": "<p>Thanks for sharing.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "447384": "Sadly, we joined in a bit late and were quite busy so we couldn't enjoy so much this very exciting competition. However, we think we did implement some things that could be interesting/useful for others. Firstly, we re-implemented paper [1] in python to extract Fourier space features from unevenly sampled data in time domain. The code can be found at [2] and it is relatively fast. It computes the magnitude, unwrapped phase, periodogram and true alarm probability of a given period to come from an actual signal. \n\n![Features for plasticc dataset with object_id 612][4]\n\nTo avoid recomputing something we have already done, we attach the hdf5 file containing Kyle's augmented dataset (thanks for sharing it by the way, it's great) together with the Fourier features [3].  This file can be read using the fold number as a key:\n\ndf = pd.read_hdf('fourier_feats.hdf', 'fold_0')\n\n[1] https://www.aanda.org/articles/aa/pdf/2012/09/aa19076-12.pdf\n\n[2] https://github.com/edumotya/pyriodogram\n\n[3] https://drive.google.com/file/d/12tbFYSIXrNX0Bb8sSBORbG4D71AwvyX4/view?usp=sharing\n\n\n  [4]: https://github.com/edumotya/pyriodogram/blob/master/example.png?raw=true",
    "448116": "Thanks for sharing."
  },
  "source": "meta"
}