{
  "id": 58470,
  "title": "How to decide in case of same volume/layer/module id?",
  "url": "/competitions/trackml-particle-identification/discussion/58470",
  "author_name": "",
  "post_date": "2018-06-08T13:25:17.500142500Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey, I wonder if/how you have achieved to pick the correct hit (from 2 candidate hits), in situations where those 2 hits are very close to each other?</p>\n\n<p>E.g. take event \"train_1/event000001005\", and look at the true track with the particle_id \"225180256246431744\". My algorithms lump (in the situation of this track) the two hits 72066 and 72067 both into 1 suggested track.</p>\n\n<p>Now, by the organizers instructions, we know that no two hits of the same track can have the same ids for volume/layer/module, and so we know that we MUST pick at most 1 of the two above hits for the suggested track (the two hits above have same volume/layer/module ids).</p>\n\n<p>But, I can not see how I can pick the correct one. My algorithms (HDBscan and Hough Transform) always lump those two hits together (they are really close).</p>\n\n<p>Any one spent some thoughts on this?</p>",
  "messages": [
    {
      "id": "340152",
      "postDate": "06/08/2018 13:25:17",
      "content": "<p>Hey, I wonder if/how you have achieved to pick the correct hit (from 2 candidate hits), in situations where those 2 hits are very close to each other?</p>\n\n<p>E.g. take event \"train_1/event000001005\", and look at the true track with the particle_id \"225180256246431744\". My algorithms lump (in the situation of this track) the two hits 72066 and 72067 both into 1 suggested track.</p>\n\n<p>Now, by the organizers instructions, we know that no two hits of the same track can have the same ids for volume/layer/module, and so we know that we MUST pick at most 1 of the two above hits for the suggested track (the two hits above have same volume/layer/module ids).</p>\n\n<p>But, I can not see how I can pick the correct one. My algorithms (HDBscan and Hough Transform) always lump those two hits together (they are really close).</p>\n\n<p>Any one spent some thoughts on this?</p>",
      "rawMarkdown": "Hey, I wonder if/how you have achieved to pick the correct hit (from 2 candidate hits), in situations where those 2 hits are very close to each other?\n\nE.g. take event \"train_1/event000001005\", and look at the true track with the particle_id \"225180256246431744\". My algorithms lump (in the situation of this track) the two hits 72066 and 72067 both into 1 suggested track.\n\nNow, by the organizers instructions, we know that no two hits of the same track can have the same ids for volume/layer/module, and so we know that we MUST pick at most 1 of the two above hits for the suggested track (the two hits above have same volume/layer/module ids).\n\nBut, I can not see how I can pick the correct one. My algorithms (HDBscan and Hough Transform) always lump those two hits together (they are really close).\n\nAny one spent some thoughts on this?",
      "votes": null
    },
    {
      "id": "340157",
      "postDate": "06/08/2018 13:36:48",
      "content": "<p>I use an algorithm that do multiple passes thru the data and I chose the hit that is closest to the mean value of the cluster's parameters. The other hit is returned to the pull to be picked up by a different cluster.    </p>",
      "rawMarkdown": "I use an algorithm that do multiple passes thru the data and I chose the hit that is closest to the mean value of the cluster's parameters. The other hit is returned to the pull to be picked up by a different cluster.",
      "votes": null
    },
    {
      "id": "340165",
      "postDate": "06/08/2018 13:57:23",
      "content": "<p>Thanks for this suggestion! I will give it a try.</p>",
      "rawMarkdown": "Thanks for this suggestion! I will give it a try.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 340157,
      "author_name": "yuval6967",
      "author_url": "",
      "post_date": "06/08/2018 13:36:48",
      "content": "<p>I use an algorithm that do multiple passes thru the data and I chose the hit that is closest to the mean value of the cluster's parameters. The other hit is returned to the pull to be picked up by a different cluster.    </p>",
      "votes": null,
      "replies": [
        {
          "id": 340165,
          "author_name": "trian2018",
          "author_url": "",
          "post_date": "06/08/2018 13:57:23",
          "content": "<p>Thanks for this suggestion! I will give it a try.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "340152": "Hey, I wonder if/how you have achieved to pick the correct hit (from 2 candidate hits), in situations where those 2 hits are very close to each other?\n\nE.g. take event \"train_1/event000001005\", and look at the true track with the particle_id \"225180256246431744\". My algorithms lump (in the situation of this track) the two hits 72066 and 72067 both into 1 suggested track.\n\nNow, by the organizers instructions, we know that no two hits of the same track can have the same ids for volume/layer/module, and so we know that we MUST pick at most 1 of the two above hits for the suggested track (the two hits above have same volume/layer/module ids).\n\nBut, I can not see how I can pick the correct one. My algorithms (HDBscan and Hough Transform) always lump those two hits together (they are really close).\n\nAny one spent some thoughts on this?",
    "340157": "I use an algorithm that do multiple passes thru the data and I chose the hit that is closest to the mean value of the cluster's parameters. The other hit is returned to the pull to be picked up by a different cluster.",
    "340165": "Thanks for this suggestion! I will give it a try."
  },
  "source": "meta"
}