{
  "id": 228377,
  "title": "Will there be a new building in the private test set?",
  "url": "/competitions/indoor-location-navigation/discussion/228377",
  "author_name": "",
  "post_date": "2021-03-24T12:34:17.943675900Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>If there is a new building in the private test set, which means there are new BSSIDs, Can we just inference the waypoint of the path just by RSSI or IMU data?<br>\nI don't think the model will work well on a new site because all the training values are relative values.</p>",
  "messages": [
    {
      "id": "1251012",
      "postDate": "03/24/2021 12:34:17",
      "content": "<p>If there is a new building in the private test set, which means there are new BSSIDs, Can we just inference the waypoint of the path just by RSSI or IMU data?<br>\nI don't think the model will work well on a new site because all the training values are relative values.</p>",
      "rawMarkdown": "If there is a new building in the private test set, which means there are new BSSIDs, Can we just inference the waypoint of the path just by RSSI or IMU data?\nI don't think the model will work well on a new site because all the training values are relative values.",
      "votes": null
    },
    {
      "id": "1252031",
      "postDate": "03/25/2021 11:19:29",
      "content": "<p>As George HW Bush might have said: \"Watch my lips, no new buildings.\" In fact, no new data. The \"private\" dataset is about 8500 lines of the existing test data, and the \"public\" dataset is about 1500 lines of it. The split seems to be done in such a way that whole paths appear in either one dataset or the other. I know that at least some of the buildings contain paths from both the public and private subsets.</p>",
      "rawMarkdown": "As George HW Bush might have said: \"Watch my lips, no new buildings.\" In fact, no new data. The \"private\" dataset is about 8500 lines of the existing test data, and the \"public\" dataset is about 1500 lines of it. The split seems to be done in such a way that whole paths appear in either one dataset or the other. I know that at least some of the buildings contain paths from both the public and private subsets.",
      "votes": null
    },
    {
      "id": "1252064",
      "postDate": "03/25/2021 11:54:36",
      "content": "<p>If so, We could use one model to fit one building, not all sites.<br>\nOr we could even use one model for one floor.</p>",
      "rawMarkdown": "If so, We could use one model to fit one building, not all sites.\nOr we could even use one model for one floor.",
      "votes": null
    },
    {
      "id": "1252091",
      "postDate": "03/25/2021 12:20:19",
      "content": "<p>Indeed, some of the leading competitors use a different model for each building.</p>",
      "rawMarkdown": "Indeed, some of the leading competitors use a different model for each building.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1252031,
      "author_name": "jbomitchell",
      "author_url": "",
      "post_date": "03/25/2021 11:19:29",
      "content": "<p>As George HW Bush might have said: \"Watch my lips, no new buildings.\" In fact, no new data. The \"private\" dataset is about 8500 lines of the existing test data, and the \"public\" dataset is about 1500 lines of it. The split seems to be done in such a way that whole paths appear in either one dataset or the other. I know that at least some of the buildings contain paths from both the public and private subsets.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1252064,
          "author_name": "xiaoyaopeng",
          "author_url": "",
          "post_date": "03/25/2021 11:54:36",
          "content": "<p>If so, We could use one model to fit one building, not all sites.<br>\nOr we could even use one model for one floor.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1252091,
          "author_name": "jbomitchell",
          "author_url": "",
          "post_date": "03/25/2021 12:20:19",
          "content": "<p>Indeed, some of the leading competitors use a different model for each building.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1251012": "If there is a new building in the private test set, which means there are new BSSIDs, Can we just inference the waypoint of the path just by RSSI or IMU data?\nI don't think the model will work well on a new site because all the training values are relative values.",
    "1252031": "As George HW Bush might have said: \"Watch my lips, no new buildings.\" In fact, no new data. The \"private\" dataset is about 8500 lines of the existing test data, and the \"public\" dataset is about 1500 lines of it. The split seems to be done in such a way that whole paths appear in either one dataset or the other. I know that at least some of the buildings contain paths from both the public and private subsets.",
    "1252064": "If so, We could use one model to fit one building, not all sites.\nOr we could even use one model for one floor.",
    "1252091": "Indeed, some of the leading competitors use a different model for each building."
  },
  "source": "meta"
}