{
  "id": 56580,
  "title": "solution for LB score 0.24",
  "url": "/competitions/trackml-particle-identification/discussion/56580",
  "author_name": "",
  "post_date": "2018-05-11T18:18:44.239235100Z",
  "votes": 25,
  "comment_count": 3,
  "views": 0,
  "content": "<p>there are mainly 2 kind of tracks:</p>\n\n<ul>\n<li><p>blue: those exit at the cap detectors</p></li>\n<li><p>red: those exit at the cylindrical detectors</p>\n\n<p>they require different parameters to detect them. You can tune the parameters of of the reference code of dbscan and hough transform (from <a href=\"https://www.kaggle.com/c/trackml-particle-identification/discussion/55708\">https://www.kaggle.com/c/trackml-particle-identification/discussion/55708</a>) to  detect them separately.</p></li>\n</ul>\n\n<p>As an example, here are my results.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/327514/9404/combine1.png\" alt=\"enter image description here\"></p>\n\n<p>Note: spend some time to draw out the results. you can discover how to tune the parameters of your algorithm by observing the results.</p>",
  "messages": [
    {
      "id": "327514",
      "postDate": "05/11/2018 18:18:44",
      "content": "<p>there are mainly 2 kind of tracks:</p>\n\n<ul>\n<li><p>blue: those exit at the cap detectors</p></li>\n<li><p>red: those exit at the cylindrical detectors</p>\n\n<p>they require different parameters to detect them. You can tune the parameters of of the reference code of dbscan and hough transform (from <a href=\"https://www.kaggle.com/c/trackml-particle-identification/discussion/55708\">https://www.kaggle.com/c/trackml-particle-identification/discussion/55708</a>) to  detect them separately.</p></li>\n</ul>\n\n<p>As an example, here are my results.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/327514/9404/combine1.png\" alt=\"enter image description here\"></p>\n\n<p>Note: spend some time to draw out the results. you can discover how to tune the parameters of your algorithm by observing the results.</p>",
      "rawMarkdown": "there are mainly 2 kind of tracks:\n\n- blue: those exit at the cap detectors\n\n- red: those exit at the cylindrical detectors\n\n they require different parameters to detect them. You can tune the parameters of of the reference code of dbscan and hough transform (from https://www.kaggle.com/c/trackml-particle-identification/discussion/55708) to  detect them separately.\n\nAs an example, here are my results.\n\n  ![enter image description here][1]\n\nNote: spend some time to draw out the results. you can discover how to tune the parameters of your algorithm by observing the results.\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/327514/9404/combine1.png",
      "votes": null
    },
    {
      "id": "328115",
      "postDate": "05/13/2018 11:49:06",
      "content": "<p>about 32% of the tracks are almost perfectly straight and they make up about 0.368 of the total metric score.</p>\n\n<p>about 43% of the track are quite straight, makeup 0.52 of metric score.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9413/straight1.png\" alt=\"enter image description here\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9414/straight.png\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "about 32% of the tracks are almost perfectly straight and they make up about 0.368 of the total metric score.\n\nabout 43% of the track are quite straight, makeup 0.52 of metric score.\n \n  ![enter image description here][1]\n\n  ![enter image description here][2]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9413/straight1.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9414/straight.png",
      "votes": null
    },
    {
      "id": "328381",
      "postDate": "05/14/2018 06:05:53",
      "content": "<p>So sliding a cone across 360 degrees on the projected x-y plane should be enough to capture these 32% tracks? </p>",
      "rawMarkdown": "So sliding a cone across 360 degrees on the projected x-y plane should be enough to capture these 32% tracks?",
      "votes": null
    },
    {
      "id": "329825",
      "postDate": "05/17/2018 08:19:29",
      "content": "<p>As in for each 'edge' of the cone where multiple points form a line to the tip, those points represent a track? For the straight tracks, would it be best to first find perfectly straight tracks, would you want to categorize before the Hough transform?</p>\n\n<p>Would the process be something like:</p>\n\n<p>Find linear tracks\nHough Transform unclassified tracks (68%)\nCluster Analysis of remaining helices </p>",
      "rawMarkdown": "As in for each 'edge' of the cone where multiple points form a line to the tip, those points represent a track? For the straight tracks, would it be best to first find perfectly straight tracks, would you want to categorize before the Hough transform?\n\nWould the process be something like:\n\nFind linear tracks\nHough Transform unclassified tracks (68%)\nCluster Analysis of remaining helices",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 328115,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "05/13/2018 11:49:06",
      "content": "<p>about 32% of the tracks are almost perfectly straight and they make up about 0.368 of the total metric score.</p>\n\n<p>about 43% of the track are quite straight, makeup 0.52 of metric score.</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9413/straight1.png\" alt=\"enter image description here\"></p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9414/straight.png\" alt=\"enter image description here\"></p>",
      "votes": null,
      "replies": [
        {
          "id": 328381,
          "author_name": "riadsouissi",
          "author_url": "",
          "post_date": "05/14/2018 06:05:53",
          "content": "<p>So sliding a cone across 360 degrees on the projected x-y plane should be enough to capture these 32% tracks? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 329825,
          "author_name": "macfarll",
          "author_url": "",
          "post_date": "05/17/2018 08:19:29",
          "content": "<p>As in for each 'edge' of the cone where multiple points form a line to the tip, those points represent a track? For the straight tracks, would it be best to first find perfectly straight tracks, would you want to categorize before the Hough transform?</p>\n\n<p>Would the process be something like:</p>\n\n<p>Find linear tracks\nHough Transform unclassified tracks (68%)\nCluster Analysis of remaining helices </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "327514": "there are mainly 2 kind of tracks:\n\n- blue: those exit at the cap detectors\n\n- red: those exit at the cylindrical detectors\n\n they require different parameters to detect them. You can tune the parameters of of the reference code of dbscan and hough transform (from https://www.kaggle.com/c/trackml-particle-identification/discussion/55708) to  detect them separately.\n\nAs an example, here are my results.\n\n  ![enter image description here][1]\n\nNote: spend some time to draw out the results. you can discover how to tune the parameters of your algorithm by observing the results.\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/327514/9404/combine1.png",
    "328115": "about 32% of the tracks are almost perfectly straight and they make up about 0.368 of the total metric score.\n\nabout 43% of the track are quite straight, makeup 0.52 of metric score.\n \n  ![enter image description here][1]\n\n  ![enter image description here][2]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9413/straight1.png\n  [2]: https://storage.googleapis.com/kaggle-forum-message-attachments/328115/9414/straight.png",
    "328381": "So sliding a cone across 360 degrees on the projected x-y plane should be enough to capture these 32% tracks?",
    "329825": "As in for each 'edge' of the cone where multiple points form a line to the tip, those points represent a track? For the straight tracks, would it be best to first find perfectly straight tracks, would you want to categorize before the Hough transform?\n\nWould the process be something like:\n\nFind linear tracks\nHough Transform unclassified tracks (68%)\nCluster Analysis of remaining helices"
  },
  "source": "meta"
}