{
  "id": 5215,
  "title": "Could you please clarify and fix these problems",
  "url": "/competitions/multi-modal-gesture-recognition/discussion/5215",
  "author_name": "",
  "post_date": "2013-07-28T01:10:25.060Z",
  "votes": null,
  "comment_count": 2,
  "views": 1121,
  "content": "<p>1) train id has four digits but test id has three digits</p>\n<p>2) in&nbsp;calcError.m the file format is</p>\n<p>SampleID,comma separated sequence of gesture identifiers</p>\n<p>but in example submission file the format is</p>\n<p>SampleID,space separated sequence of gesture identifiers</p>\n<p>3)&nbsp;calcError.m doesn't produce similar score as the leaderboard. I think some normalization should be done in leaderboard scores. Could you please make the formula clear?</p>\n<p>4)&nbsp;Sample00100 doesn't have label. It isn't necessary to include this file.</p>\n<p>5) is the number of gestures fixed as 10 for test data?</p>\n<p>6) the deadline shown in the top-right of the page doesn't seem to be correct.</p>\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "27724",
      "postDate": "07/28/2013 01:10:25",
      "content": "<p>1) train id has four digits but test id has three digits</p>\n<p>2) in&nbsp;calcError.m the file format is</p>\n<p>SampleID,comma separated sequence of gesture identifiers</p>\n<p>but in example submission file the format is</p>\n<p>SampleID,space separated sequence of gesture identifiers</p>\n<p>3)&nbsp;calcError.m doesn't produce similar score as the leaderboard. I think some normalization should be done in leaderboard scores. Could you please make the formula clear?</p>\n<p>4)&nbsp;Sample00100 doesn't have label. It isn't necessary to include this file.</p>\n<p>5) is the number of gestures fixed as 10 for test data?</p>\n<p>6) the deadline shown in the top-right of the page doesn't seem to be correct.</p>\n<p>Thanks!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "27725",
      "postDate": "07/28/2013 05:10:58",
      "content": "<p>I also had a question about the validation dataset: the random sample csv file contains exactly 20 gesture annotations for each sample - is that known (i.e. each validation video contains 20 gestures - with possible repetitions)? Because I looked at the first two videos in the validation dataset (410 and 411) and the both seemed to contain less gestures than 20.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "27837",
      "postDate": "07/30/2013 14:35:20",
      "content": "<p>Dear all,</p>\n<p>the format to submit the results to the Kaggle platform is:</p>\n<p>&lt;SampleID&gt;,&lt;space separated gesture - prediction&gt;</p>\n<p>where the SampleID (column Id) is parsed as an integer, therefore, it doesn't mater if it is 3 or 4 digits long. </p>\n<p>The provided Matlab script uses Levenshtain edition distance between the prediction and the ground truth, but is not the same script that is used in the Kaggle platform. We are checking the script in order to provide a better approximation. Nevertheless, we encourage you to use the Kaggle platform to validate your predictions, which is the metric that will be used to rank the final predictions.</p>\n<p>We reported that some files had problems, the case of Sample00100, the labels are missing and other few samples has problems with the exportation, therefore some information was missing. We are doing extensive validation over final evaluation data to avoid any of those problems.</p>\n<p>Regarding the number of gestures, it is not fixed, and is part of the difficulty of the challenge. Most of the training samples have all 20 gestures, but the rest, and most of the validation samples have noisy gestures (gestures different from the 20 target gestures) and a more reduced number of target gestures (generally between 9 and 12). Noisy gestures should not be present in the final predicted sequence.</p>\n<p>The correct deadline is the one showed at the &quot;timeline&quot; section, therefore, August the 20th.&nbsp;</p>\n<p>&nbsp;</p>\n<p>I hope that this entry clarify some of your doubts.</p>\n<p>&nbsp;</p>\n<p>Thanks for your interest.</p>\n<p>&nbsp;</p>\n<p>Xavier</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 27725,
      "author_name": "onepoint",
      "author_url": "",
      "post_date": "07/28/2013 05:10:58",
      "content": "<p>I also had a question about the validation dataset: the random sample csv file contains exactly 20 gesture annotations for each sample - is that known (i.e. each validation video contains 20 gestures - with possible repetitions)? Because I looked at the first two videos in the validation dataset (410 and 411) and the both seemed to contain less gestures than 20.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 27837,
      "author_name": "xbaro100272",
      "author_url": "",
      "post_date": "07/30/2013 14:35:20",
      "content": "<p>Dear all,</p>\n<p>the format to submit the results to the Kaggle platform is:</p>\n<p>&lt;SampleID&gt;,&lt;space separated gesture - prediction&gt;</p>\n<p>where the SampleID (column Id) is parsed as an integer, therefore, it doesn't mater if it is 3 or 4 digits long. </p>\n<p>The provided Matlab script uses Levenshtain edition distance between the prediction and the ground truth, but is not the same script that is used in the Kaggle platform. We are checking the script in order to provide a better approximation. Nevertheless, we encourage you to use the Kaggle platform to validate your predictions, which is the metric that will be used to rank the final predictions.</p>\n<p>We reported that some files had problems, the case of Sample00100, the labels are missing and other few samples has problems with the exportation, therefore some information was missing. We are doing extensive validation over final evaluation data to avoid any of those problems.</p>\n<p>Regarding the number of gestures, it is not fixed, and is part of the difficulty of the challenge. Most of the training samples have all 20 gestures, but the rest, and most of the validation samples have noisy gestures (gestures different from the 20 target gestures) and a more reduced number of target gestures (generally between 9 and 12). Noisy gestures should not be present in the final predicted sequence.</p>\n<p>The correct deadline is the one showed at the &quot;timeline&quot; section, therefore, August the 20th.&nbsp;</p>\n<p>&nbsp;</p>\n<p>I hope that this entry clarify some of your doubts.</p>\n<p>&nbsp;</p>\n<p>Thanks for your interest.</p>\n<p>&nbsp;</p>\n<p>Xavier</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "27724": "",
    "27725": "",
    "27837": ""
  },
  "source": "meta"
}