{
  "id": 583568,
  "title": "Sequence (SEQ_011975) with only Transition phase (or Transition + Pause)",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/583568",
  "author_name": "",
  "post_date": "2025-06-07T21:43:08.893302400Z",
  "votes": 29,
  "comment_count": 1,
  "views": 0,
  "content": "<p>From the <a href=\"https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/data\" target=\"_blank\">data page</a> we know that \"Each sequence includes one Transition, one Pause, and one Gesture.\" </p>\n<p>However \"Phase\" column has only two values - \"Transition\" and \"Gesture\"</p>\n<p>It seems this is a typical pattern </p>\n<ol>\n<li><strong>Transition phase</strong>: 'Moves hand to target location' or 'Relaxes and moves hand to target location'</li>\n<li><strong>Transition phase</strong>: 'Hand at target location' (this is the \"Pause\") - Also mentioned <a href=\"https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583357#3218481\" target=\"_blank\">here</a></li>\n<li><strong>Gesture phase</strong>: 'Performs gesture'</li>\n</ol>\n<p>We have 8151 unique sequences - All of them have both \"Transition\" and \"Gesture\" phases, except for SEQ_011975, which only has the \"Transition\" phase.</p>\n<p>Not sure why - perhaps missing data/outlier/not to be included in training.</p>",
  "messages": [
    {
      "id": "3219519",
      "postDate": "06/07/2025 21:43:08",
      "content": "<p>From the <a href=\"https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/data\" target=\"_blank\">data page</a> we know that \"Each sequence includes one Transition, one Pause, and one Gesture.\" </p>\n<p>However \"Phase\" column has only two values - \"Transition\" and \"Gesture\"</p>\n<p>It seems this is a typical pattern </p>\n<ol>\n<li><strong>Transition phase</strong>: 'Moves hand to target location' or 'Relaxes and moves hand to target location'</li>\n<li><strong>Transition phase</strong>: 'Hand at target location' (this is the \"Pause\") - Also mentioned <a href=\"https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583357#3218481\" target=\"_blank\">here</a></li>\n<li><strong>Gesture phase</strong>: 'Performs gesture'</li>\n</ol>\n<p>We have 8151 unique sequences - All of them have both \"Transition\" and \"Gesture\" phases, except for SEQ_011975, which only has the \"Transition\" phase.</p>\n<p>Not sure why - perhaps missing data/outlier/not to be included in training.</p>",
      "rawMarkdown": "From the [data page](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/data) we know that \"Each sequence includes one Transition, one Pause, and one Gesture.\" \n\nHowever \"Phase\" column has only two values - \"Transition\" and \"Gesture\"\n\nIt seems this is a typical pattern \n1. **Transition phase**: 'Moves hand to target location' or 'Relaxes and moves hand to target location'\n2. **Transition phase**: 'Hand at target location' (this is the \"Pause\") - Also mentioned [here](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583357#3218481)\n3. **Gesture phase**: 'Performs gesture'\n\nWe have 8151 unique sequences - All of them have both \"Transition\" and \"Gesture\" phases, except for SEQ_011975, which only has the \"Transition\" phase.\n\n\n\nNot sure why - perhaps missing data/outlier/not to be included in training.",
      "votes": null
    },
    {
      "id": "3219578",
      "postDate": "06/08/2025 02:05:02",
      "content": "<p>I tried to look into this a little. SEQ_011975 is subject SUBJ_030676 performing the \"Text on phone\" gesture while lying on their side.<br>\nIn the training data, this subject has 7 other recordings for this gesture, but only 1 of these is while lying on their side: SEQ_053059.</p>\n<p>Below, I plotted each of the gestures over time and marked the phase changes:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8509053%2F3f19637b0bfd08dac4fdc22424bfc605%2Fplot_screenshot_fix.png?generation=1749348568791932&amp;alt=media\" alt=\"\"><br>\nSEQ_011975 is much shorter than SEQ_053059 (also lying on side), this could suggest the end of the sequence didn't make it into the dataset? There are other short sequences though, so another explanation could be that the phase change just wasn't recorded correctly.</p>\n<p>I think for now it makes sense to remove this sequence from our training data, but maybe there's a way to get a better idea of what happened from comparing the IMU data between sequences. Thanks for drawing attention to this!</p>",
      "rawMarkdown": "I tried to look into this a little. SEQ_011975 is subject SUBJ_030676 performing the \"Text on phone\" gesture while lying on their side.\nIn the training data, this subject has 7 other recordings for this gesture, but only 1 of these is while lying on their side: SEQ_053059.\n\nBelow, I plotted each of the gestures over time and marked the phase changes:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8509053%2F3f19637b0bfd08dac4fdc22424bfc605%2Fplot_screenshot_fix.png?generation=1749348568791932&alt=media)\nSEQ_011975 is much shorter than SEQ_053059 (also lying on side), this could suggest the end of the sequence didn't make it into the dataset? There are other short sequences though, so another explanation could be that the phase change just wasn't recorded correctly.\n\nI think for now it makes sense to remove this sequence from our training data, but maybe there's a way to get a better idea of what happened from comparing the IMU data between sequences. Thanks for drawing attention to this!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3219578,
      "author_name": "petermoorhouse",
      "author_url": "",
      "post_date": "06/08/2025 02:05:02",
      "content": "<p>I tried to look into this a little. SEQ_011975 is subject SUBJ_030676 performing the \"Text on phone\" gesture while lying on their side.<br>\nIn the training data, this subject has 7 other recordings for this gesture, but only 1 of these is while lying on their side: SEQ_053059.</p>\n<p>Below, I plotted each of the gestures over time and marked the phase changes:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8509053%2F3f19637b0bfd08dac4fdc22424bfc605%2Fplot_screenshot_fix.png?generation=1749348568791932&amp;alt=media\" alt=\"\"><br>\nSEQ_011975 is much shorter than SEQ_053059 (also lying on side), this could suggest the end of the sequence didn't make it into the dataset? There are other short sequences though, so another explanation could be that the phase change just wasn't recorded correctly.</p>\n<p>I think for now it makes sense to remove this sequence from our training data, but maybe there's a way to get a better idea of what happened from comparing the IMU data between sequences. Thanks for drawing attention to this!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3219519": "From the [data page](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/data) we know that \"Each sequence includes one Transition, one Pause, and one Gesture.\" \n\nHowever \"Phase\" column has only two values - \"Transition\" and \"Gesture\"\n\nIt seems this is a typical pattern \n1. **Transition phase**: 'Moves hand to target location' or 'Relaxes and moves hand to target location'\n2. **Transition phase**: 'Hand at target location' (this is the \"Pause\") - Also mentioned [here](https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583357#3218481)\n3. **Gesture phase**: 'Performs gesture'\n\nWe have 8151 unique sequences - All of them have both \"Transition\" and \"Gesture\" phases, except for SEQ_011975, which only has the \"Transition\" phase.\n\n\n\nNot sure why - perhaps missing data/outlier/not to be included in training.",
    "3219578": "I tried to look into this a little. SEQ_011975 is subject SUBJ_030676 performing the \"Text on phone\" gesture while lying on their side.\nIn the training data, this subject has 7 other recordings for this gesture, but only 1 of these is while lying on their side: SEQ_053059.\n\nBelow, I plotted each of the gestures over time and marked the phase changes:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8509053%2F3f19637b0bfd08dac4fdc22424bfc605%2Fplot_screenshot_fix.png?generation=1749348568791932&alt=media)\nSEQ_011975 is much shorter than SEQ_053059 (also lying on side), this could suggest the end of the sequence didn't make it into the dataset? There are other short sequences though, so another explanation could be that the phase change just wasn't recorded correctly.\n\nI think for now it makes sense to remove this sequence from our training data, but maybe there's a way to get a better idea of what happened from comparing the IMU data between sequences. Thanks for drawing attention to this!"
  },
  "source": "meta"
}