{
  "id": 181324,
  "title": "Question about timestamps",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/181324",
  "author_name": "Snihyr Kostiantyn",
  "post_date": "2020-09-08T12:43:11.361000",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I have an item from <code>AgentDataset</code> not really understand nature of <code>timestamp</code> field.<br>\nLet's fetch agents for some scene with <code>SCENE_IDX</code> index.<br>\nAfter let's check scene time bounds.</p>\n<pre><code>SCENE_IDX = 74\nagents = dataset.get_scene_dataset(SCENE_IDX)\ndf = pd.DataFrame(iter(agents))\nscene_record = zarr_dataset.scenes[SCENE_IDX]\nscene_record['start_time'] - df['timestamp'].min(), scene_record['end_time'] - df['timestamp'].max()\n</code></pre>\n<p>Output is <br>\n<code>(-1153141166, 147435066)</code></p>\n<p>Seems like response dataset contains agents with timestamps less than scene start time (like, trajectory started in previous scene )  Should not they have to had prev scene positions in <code>history_positions</code> (they are [[0.0, 0.0]]) ?<br>\nUpd: tagged <a href=\"https://www.kaggle.com/iglovikov\" target=\"_blank\">@iglovikov</a> as a host</p>",
  "messages": [
    {
      "id": 1002803,
      "postDate": "2020-09-08T12:43:11.363Z",
      "content": "<p>I have an item from <code>AgentDataset</code> not really understand nature of <code>timestamp</code> field.<br>\nLet's fetch agents for some scene with <code>SCENE_IDX</code> index.<br>\nAfter let's check scene time bounds.</p>\n<pre><code>SCENE_IDX = 74\nagents = dataset.get_scene_dataset(SCENE_IDX)\ndf = pd.DataFrame(iter(agents))\nscene_record = zarr_dataset.scenes[SCENE_IDX]\nscene_record['start_time'] - df['timestamp'].min(), scene_record['end_time'] - df['timestamp'].max()\n</code></pre>\n<p>Output is <br>\n<code>(-1153141166, 147435066)</code></p>\n<p>Seems like response dataset contains agents with timestamps less than scene start time (like, trajectory started in previous scene )  Should not they have to had prev scene positions in <code>history_positions</code> (they are [[0.0, 0.0]]) ?<br>\nUpd: tagged <a href=\"https://www.kaggle.com/iglovikov\" target=\"_blank\">@iglovikov</a> as a host</p>",
      "rawMarkdown": "I have an item from `AgentDataset` not really understand nature of `timestamp` field.\nLet's fetch agents for some scene with `SCENE_IDX` index.\nAfter let's check scene time bounds.\n```\nSCENE_IDX = 74\nagents = dataset.get_scene_dataset(SCENE_IDX)\ndf = pd.DataFrame(iter(agents))\nscene_record = zarr_dataset.scenes[SCENE_IDX]\nscene_record['start_time'] - df['timestamp'].min(), scene_record['end_time'] - df['timestamp'].max()\n```\n\nOutput is \n`(-1153141166, 147435066)`\n\nSeems like response dataset contains agents with timestamps less than scene start time (like, trajectory started in previous scene )  Should not they have to had prev scene positions in `history_positions` (they are [[0.0, 0.0]]) ?\nUpd: tagged @iglovikov as a host",
      "votes": 1
    },
    {
      "id": 1003090,
      "postDate": "2020-09-08T16:47:51.983Z",
      "content": "<p>Hey, can I ask you which set are you working on (test, train, etc..)? That way I can check it thoroughly :) </p>",
      "rawMarkdown": "Hey, can I ask you which set are you working on (test, train, etc..)? That way I can check it thoroughly :) ",
      "replies": [
        {
          "id": 1003807,
          "postDate": "2020-09-09T09:06:58.430Z",
          "content": "<p>It's the sample set. <a href=\"https://www.kaggle.com/lexplua/timestamps-in-lyft-dataset\" target=\"_blank\">https://www.kaggle.com/lexplua/timestamps-in-lyft-dataset</a> here's notebook<br>\nupd: first time sharing notebook, added you as collaborator if link doesn't work</p>",
          "rawMarkdown": "It's the sample set. https://www.kaggle.com/lexplua/timestamps-in-lyft-dataset here's notebook\nupd: first time sharing notebook, added you as collaborator if link doesn't work"
        },
        {
          "id": 1004092,
          "postDate": "2020-09-09T12:59:44.383Z",
          "content": "<p>cool, I'll take a look into it later today :) </p>",
          "rawMarkdown": "cool, I'll take a look into it later today :) ",
          "votes": 1
        },
        {
          "id": 1004396,
          "postDate": "2020-09-09T17:30:08.877Z",
          "content": "<p>Actually, after thinking a bit about it results look fine to me:</p>\n<p>you're doing <code>start_time - min(timestamps)</code> which yields a negative number. This means the smallest timestamps is still greater than the <code>start_time</code> of the scene right? a timestamp outside the scene boundaries (i.e. before <code>start_time</code>) would give you a positive number.</p>\n<p>To summarise, <code>timestamps</code> in a scene should always be in the <code>(start_time, end_time)</code> range. Does this make sense to you?</p>",
          "rawMarkdown": "Actually, after thinking a bit about it results look fine to me:\n\nyou're doing `start_time - min(timestamps)` which yields a negative number. This means the smallest timestamps is still greater than the `start_time` of the scene right? a timestamp outside the scene boundaries (i.e. before `start_time`) would give you a positive number.\n\nTo summarise, `timestamps` in a scene should always be in the `(start_time, end_time)` range. Does this make sense to you?",
          "votes": 1
        },
        {
          "id": 1005101,
          "postDate": "2020-09-10T08:52:20.657Z",
          "content": "<p>Oh, yes, sure, sorry</p>",
          "rawMarkdown": "Oh, yes, sure, sorry"
        }
      ]
    },
    {
      "id": 1003803,
      "postDate": "2020-09-09T09:06:44.143Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1003090,
      "author_name": "Luca Bergamini",
      "author_url": "",
      "post_date": "2020-09-08T16:47:51.983000",
      "content": "<p>Hey, can I ask you which set are you working on (test, train, etc..)? That way I can check it thoroughly :) </p>",
      "votes": 0,
      "replies": [
        {
          "id": 1003807,
          "author_name": "Snihyr Kostiantyn",
          "author_url": "",
          "post_date": "2020-09-09T09:06:58.430000",
          "content": "<p>It's the sample set. <a href=\"https://www.kaggle.com/lexplua/timestamps-in-lyft-dataset\" target=\"_blank\">https://www.kaggle.com/lexplua/timestamps-in-lyft-dataset</a> here's notebook<br>\nupd: first time sharing notebook, added you as collaborator if link doesn't work</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1004092,
          "author_name": "Luca Bergamini",
          "author_url": "",
          "post_date": "2020-09-09T12:59:44.383000",
          "content": "<p>cool, I'll take a look into it later today :) </p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1004396,
          "author_name": "Luca Bergamini",
          "author_url": "",
          "post_date": "2020-09-09T17:30:08.877000",
          "content": "<p>Actually, after thinking a bit about it results look fine to me:</p>\n<p>you're doing <code>start_time - min(timestamps)</code> which yields a negative number. This means the smallest timestamps is still greater than the <code>start_time</code> of the scene right? a timestamp outside the scene boundaries (i.e. before <code>start_time</code>) would give you a positive number.</p>\n<p>To summarise, <code>timestamps</code> in a scene should always be in the <code>(start_time, end_time)</code> range. Does this make sense to you?</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1005101,
          "author_name": "Snihyr Kostiantyn",
          "author_url": "",
          "post_date": "2020-09-10T08:52:20.657000",
          "content": "<p>Oh, yes, sure, sorry</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1003803,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-09-09T09:06:44.143000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1002803": "I have an item from `AgentDataset` not really understand nature of `timestamp` field.\nLet's fetch agents for some scene with `SCENE_IDX` index.\nAfter let's check scene time bounds.\n```\nSCENE_IDX = 74\nagents = dataset.get_scene_dataset(SCENE_IDX)\ndf = pd.DataFrame(iter(agents))\nscene_record = zarr_dataset.scenes[SCENE_IDX]\nscene_record['start_time'] - df['timestamp'].min(), scene_record['end_time'] - df['timestamp'].max()\n```\n\nOutput is \n`(-1153141166, 147435066)`\n\nSeems like response dataset contains agents with timestamps less than scene start time (like, trajectory started in previous scene )  Should not they have to had prev scene positions in `history_positions` (they are [[0.0, 0.0]]) ?\nUpd: tagged @iglovikov as a host",
    "1003090": "Hey, can I ask you which set are you working on (test, train, etc..)? That way I can check it thoroughly :) ",
    "1003803": ""
  }
}