{
  "id": 58078,
  "title": "Flattening or unrolling tracks in polar (r,s,c) coordinates",
  "url": "/competitions/trackml-particle-identification/discussion/58078",
  "author_name": "",
  "post_date": "2018-06-01T23:26:49.368675Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi,</p>\n\n<p>As mentioned in several discussion posts, transforming hits from Cartesian coordinates (x,y,x) to Polar coordinates (r,s,c) seems to be a promising way to preprocess the data. One problem that emerges after doing this is how to transform the data, now in the shape of a cylinder, to a rectangle so that it can be sent to a CNN for clustering. </p>\n\n<p>One way of doing this is to flatten the cylinder by using the rs plane. This might create problems if there are two tracks on opposite sides of the cylinder which would appear on the rs plane as one track as shown in the below.</p>\n\n<p>My question is, would it be better to 'unroll' the cylinder and if so, how would it be done?</p>\n\n<p>Thanks   </p>\n\n<p>ps:</p>\n\n<p>Hits in polar coordinates:\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9557/cartesian-to-polar-coordinates.png\" alt=\"hits in polar coordinates\"></p>\n\n<p>rs plane of hits in polar coordinates:\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9558/cartesian-to-polar-coordinates-flattened-2.jpg\" alt=\"rs plane of hits in polar coordinates\"></p>",
  "messages": [
    {
      "id": "337098",
      "postDate": "06/01/2018 23:26:49",
      "content": "<p>Hi,</p>\n\n<p>As mentioned in several discussion posts, transforming hits from Cartesian coordinates (x,y,x) to Polar coordinates (r,s,c) seems to be a promising way to preprocess the data. One problem that emerges after doing this is how to transform the data, now in the shape of a cylinder, to a rectangle so that it can be sent to a CNN for clustering. </p>\n\n<p>One way of doing this is to flatten the cylinder by using the rs plane. This might create problems if there are two tracks on opposite sides of the cylinder which would appear on the rs plane as one track as shown in the below.</p>\n\n<p>My question is, would it be better to 'unroll' the cylinder and if so, how would it be done?</p>\n\n<p>Thanks   </p>\n\n<p>ps:</p>\n\n<p>Hits in polar coordinates:\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9557/cartesian-to-polar-coordinates.png\" alt=\"hits in polar coordinates\"></p>\n\n<p>rs plane of hits in polar coordinates:\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9558/cartesian-to-polar-coordinates-flattened-2.jpg\" alt=\"rs plane of hits in polar coordinates\"></p>",
      "rawMarkdown": "Hi,\n\nAs mentioned in several discussion posts, transforming hits from Cartesian coordinates (x,y,x) to Polar coordinates (r,s,c) seems to be a promising way to preprocess the data. One problem that emerges after doing this is how to transform the data, now in the shape of a cylinder, to a rectangle so that it can be sent to a CNN for clustering. \n\nOne way of doing this is to flatten the cylinder by using the rs plane. This might create problems if there are two tracks on opposite sides of the cylinder which would appear on the rs plane as one track as shown in the below.\n\nMy question is, would it be better to 'unroll' the cylinder and if so, how would it be done?\n\nThanks   \n\nps:\n\nHits in polar coordinates:\n![hits in polar coordinates][1]\n\nrs plane of hits in polar coordinates:\n![rs plane of hits in polar coordinates][2]\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9557/cartesian-to-polar-coordinates.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9558/cartesian-to-polar-coordinates-flattened-2.jpg",
      "votes": null
    },
    {
      "id": "337285",
      "postDate": "06/02/2018 12:55:03",
      "content": "<p>If you just plot the angle instead of the sine and cosine you will have the “unrolled” version.  The problem (as pointed at in another discussion) is due to the discontinuity between 0 and 2pi you will lose a few tracks or clusters. </p>",
      "rawMarkdown": "If you just plot the angle instead of the sine and cosine you will have the “unrolled” version.  The problem (as pointed at in another discussion) is due to the discontinuity between 0 and 2pi you will lose a few tracks or clusters.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 337285,
      "author_name": "dmriser",
      "author_url": "",
      "post_date": "06/02/2018 12:55:03",
      "content": "<p>If you just plot the angle instead of the sine and cosine you will have the “unrolled” version.  The problem (as pointed at in another discussion) is due to the discontinuity between 0 and 2pi you will lose a few tracks or clusters. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "337098": "Hi,\n\nAs mentioned in several discussion posts, transforming hits from Cartesian coordinates (x,y,x) to Polar coordinates (r,s,c) seems to be a promising way to preprocess the data. One problem that emerges after doing this is how to transform the data, now in the shape of a cylinder, to a rectangle so that it can be sent to a CNN for clustering. \n\nOne way of doing this is to flatten the cylinder by using the rs plane. This might create problems if there are two tracks on opposite sides of the cylinder which would appear on the rs plane as one track as shown in the below.\n\nMy question is, would it be better to 'unroll' the cylinder and if so, how would it be done?\n\nThanks   \n\nps:\n\nHits in polar coordinates:\n![hits in polar coordinates][1]\n\nrs plane of hits in polar coordinates:\n![rs plane of hits in polar coordinates][2]\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9557/cartesian-to-polar-coordinates.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/337098/9558/cartesian-to-polar-coordinates-flattened-2.jpg",
    "337285": "If you just plot the angle instead of the sine and cosine you will have the “unrolled” version.  The problem (as pointed at in another discussion) is due to the discontinuity between 0 and 2pi you will lose a few tracks or clusters."
  },
  "source": "meta"
}