{
  "id": 238331,
  "title": "Code environmental data",
  "url": "/competitions/cse151b-spring/discussion/238331",
  "author_name": "",
  "post_date": "2021-05-11T23:46:26.902125400Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi, I want to open a discussion about how to code environmental data as inputs of the deep learning model. It is easy to code vehicle data. You only need to input the position/velocity of each timestamp. However, environmental data is a bit difficult. I will share some of my ideas and you are welcome to share your ideas in the reply too.</p>",
  "messages": [
    {
      "id": "1303142",
      "postDate": "05/11/2021 23:46:26",
      "content": "<p>Hi, I want to open a discussion about how to code environmental data as inputs of the deep learning model. It is easy to code vehicle data. You only need to input the position/velocity of each timestamp. However, environmental data is a bit difficult. I will share some of my ideas and you are welcome to share your ideas in the reply too.</p>",
      "rawMarkdown": "Hi, I want to open a discussion about how to code environmental data as inputs of the deep learning model. It is easy to code vehicle data. You only need to input the position/velocity of each timestamp. However, environmental data is a bit difficult. I will share some of my ideas and you are welcome to share your ideas in the reply too.",
      "votes": null
    },
    {
      "id": "1303154",
      "postDate": "05/11/2021 23:58:46",
      "content": "<p>One of the methods I use is to have a grid of n*n. The index of each grid node represents the positional information. This way, we basically draw a 2D map with the grid. </p>\n<p>Advantages:</p>\n<ol>\n<li>The index contains the positional information and the content can contain other information.</li>\n<li>This way we can use CNN on the grid.</li>\n</ol>\n<p>Disadvantages:</p>\n<ol>\n<li>This is very space-consuming, with a lot of zeros in the grids.</li>\n<li>The speed can vary in the dataset, so the target vehicle may drive outside of your grid.</li>\n<li>The index of the grid cannot carry specific information about the position, only an estimation.</li>\n</ol>",
      "rawMarkdown": "One of the methods I use is to have a grid of n*n. The index of each grid node represents the positional information. This way, we basically draw a 2D map with the grid. \n\nAdvantages:\n1. The index contains the positional information and the content can contain other information.\n2. This way we can use CNN on the grid.\n\nDisadvantages:\n1. This is very space-consuming, with a lot of zeros in the grids.\n2. The speed can vary in the dataset, so the target vehicle may drive outside of your grid.\n3. The index of the grid cannot carry specific information about the position, only an estimation.",
      "votes": null
    },
    {
      "id": "1310575",
      "postDate": "05/16/2021 18:24:29",
      "content": "<p>I think it should be possible to make an array large enough to encode the information you're asking for. </p>\n<p>For example, if you're only using the positional data, then you'd have an input size of 120 (60 vehicles * 2 coord dims) So why no make your input size 123 (or larger/smaller) and then say arr[:3] = norm/whatever, arr[3:] = p_in?</p>",
      "rawMarkdown": "I think it should be possible to make an array large enough to encode the information you're asking for. \n\nFor example, if you're only using the positional data, then you'd have an input size of 120 (60 vehicles * 2 coord dims) So why no make your input size 123 (or larger/smaller) and then say arr[:3] = norm/whatever, arr[3:] = p_in?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1303154,
      "author_name": "felixzhang99",
      "author_url": "",
      "post_date": "05/11/2021 23:58:46",
      "content": "<p>One of the methods I use is to have a grid of n*n. The index of each grid node represents the positional information. This way, we basically draw a 2D map with the grid. </p>\n<p>Advantages:</p>\n<ol>\n<li>The index contains the positional information and the content can contain other information.</li>\n<li>This way we can use CNN on the grid.</li>\n</ol>\n<p>Disadvantages:</p>\n<ol>\n<li>This is very space-consuming, with a lot of zeros in the grids.</li>\n<li>The speed can vary in the dataset, so the target vehicle may drive outside of your grid.</li>\n<li>The index of the grid cannot carry specific information about the position, only an estimation.</li>\n</ol>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1310575,
      "author_name": "yishaisilver",
      "author_url": "",
      "post_date": "05/16/2021 18:24:29",
      "content": "<p>I think it should be possible to make an array large enough to encode the information you're asking for. </p>\n<p>For example, if you're only using the positional data, then you'd have an input size of 120 (60 vehicles * 2 coord dims) So why no make your input size 123 (or larger/smaller) and then say arr[:3] = norm/whatever, arr[3:] = p_in?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1303142": "Hi, I want to open a discussion about how to code environmental data as inputs of the deep learning model. It is easy to code vehicle data. You only need to input the position/velocity of each timestamp. However, environmental data is a bit difficult. I will share some of my ideas and you are welcome to share your ideas in the reply too.",
    "1303154": "One of the methods I use is to have a grid of n*n. The index of each grid node represents the positional information. This way, we basically draw a 2D map with the grid. \n\nAdvantages:\n1. The index contains the positional information and the content can contain other information.\n2. This way we can use CNN on the grid.\n\nDisadvantages:\n1. This is very space-consuming, with a lot of zeros in the grids.\n2. The speed can vary in the dataset, so the target vehicle may drive outside of your grid.\n3. The index of the grid cannot carry specific information about the position, only an estimation.",
    "1310575": "I think it should be possible to make an array large enough to encode the information you're asking for. \n\nFor example, if you're only using the positional data, then you'd have an input size of 120 (60 vehicles * 2 coord dims) So why no make your input size 123 (or larger/smaller) and then say arr[:3] = norm/whatever, arr[3:] = p_in?"
  },
  "source": "meta"
}