{
  "id": 180481,
  "title": "Is the time between frames the same in training and test?",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/180481",
  "author_name": "",
  "post_date": "2020-09-05T08:09:38.846771100Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>If you compute the time between frames on the training and test sets, I get 1x10^8 (nanoseconds I'm guessing?) for the training set while 2.5x10^8 for the test set. I computed this from the information given for a single scene which has the starting and ending frame index and also the starting and ending time. </p>\n<p>It is also possible that 150 of the 250 original frames were excluded from the scene when constructing the test set and the end time was not adjusted to account for this deletion. I would bet this is what happened. It would be odd if the time difference between frames were different across sets.</p>",
  "messages": [
    {
      "id": "998945",
      "postDate": "09/05/2020 08:09:38",
      "content": "<p>If you compute the time between frames on the training and test sets, I get 1x10^8 (nanoseconds I'm guessing?) for the training set while 2.5x10^8 for the test set. I computed this from the information given for a single scene which has the starting and ending frame index and also the starting and ending time. </p>\n<p>It is also possible that 150 of the 250 original frames were excluded from the scene when constructing the test set and the end time was not adjusted to account for this deletion. I would bet this is what happened. It would be odd if the time difference between frames were different across sets.</p>",
      "rawMarkdown": "If you compute the time between frames on the training and test sets, I get 1x10^8 (nanoseconds I'm guessing?) for the training set while 2.5x10^8 for the test set. I computed this from the information given for a single scene which has the starting and ending frame index and also the starting and ending time. \n\nIt is also possible that 150 of the 250 original frames were excluded from the scene when constructing the test set and the end time was not adjusted to account for this deletion. I would bet this is what happened. It would be odd if the time difference between frames were different across sets.",
      "votes": null
    },
    {
      "id": "999605",
      "postDate": "09/05/2020 19:41:26",
      "content": "<p>Test set was created using the l5kit.evaluation.create_chopped_dataset method. You could try to reproduce the behavior by using the method on the train set. Or you might study the code to see what it does to the time.<br>\nSee here for reference: <a href=\"https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset\" target=\"_blank\">https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset</a></p>\n<p>edit: checking the code, it does seem like the timestamps are not updated and only the frames are removed (100+) from each scene.</p>",
      "rawMarkdown": "Test set was created using the l5kit.evaluation.create_chopped_dataset method. You could try to reproduce the behavior by using the method on the train set. Or you might study the code to see what it does to the time.\nSee here for reference: https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset\n\nedit: checking the code, it does seem like the timestamps are not updated and only the frames are removed (100+) from each scene.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 999605,
      "author_name": "ilu000",
      "author_url": "",
      "post_date": "09/05/2020 19:41:26",
      "content": "<p>Test set was created using the l5kit.evaluation.create_chopped_dataset method. You could try to reproduce the behavior by using the method on the train set. Or you might study the code to see what it does to the time.<br>\nSee here for reference: <a href=\"https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset\" target=\"_blank\">https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset</a></p>\n<p>edit: checking the code, it does seem like the timestamps are not updated and only the frames are removed (100+) from each scene.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "998945": "If you compute the time between frames on the training and test sets, I get 1x10^8 (nanoseconds I'm guessing?) for the training set while 2.5x10^8 for the test set. I computed this from the information given for a single scene which has the starting and ending frame index and also the starting and ending time. \n\nIt is also possible that 150 of the 250 original frames were excluded from the scene when constructing the test set and the end time was not adjusted to account for this deletion. I would bet this is what happened. It would be odd if the time difference between frames were different across sets.",
    "999605": "Test set was created using the l5kit.evaluation.create_chopped_dataset method. You could try to reproduce the behavior by using the method on the train set. Or you might study the code to see what it does to the time.\nSee here for reference: https://lyft.github.io/l5kit/API/l5kit.evaluation.html#l5kit.evaluation.create_chopped_dataset\n\nedit: checking the code, it does seem like the timestamps are not updated and only the frames are removed (100+) from each scene."
  },
  "source": "meta"
}