{
  "id": 401504,
  "title": "Just one timeseries in the de dataset with a valid StartHesition event makes model selection and cross-validation difficult",
  "url": "/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/401504",
  "author_name": "",
  "post_date": "2023-04-13T15:21:55.817723800Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>From the competition guide we know that correctly annotated measures from the <code>de</code> dataset are the ones where both <code>Valid</code> and <code>Task</code> are <code>True</code>.<br>\nMaking some stats I found that with such conditions just one time series (<code>0d7ab3a9f9</code>) in the de dataset have some (88) measures where <code>StartHesitation == 1</code>.<br>\nThis makes model selection and cross-validation quite tricky.<br>\nHow are you handling this problem?</p>",
  "messages": [
    {
      "id": "2220654",
      "postDate": "04/13/2023 15:21:55",
      "content": "<p>From the competition guide we know that correctly annotated measures from the <code>de</code> dataset are the ones where both <code>Valid</code> and <code>Task</code> are <code>True</code>.<br>\nMaking some stats I found that with such conditions just one time series (<code>0d7ab3a9f9</code>) in the de dataset have some (88) measures where <code>StartHesitation == 1</code>.<br>\nThis makes model selection and cross-validation quite tricky.<br>\nHow are you handling this problem?</p>",
      "rawMarkdown": "From the competition guide we know that correctly annotated measures from the `de` dataset are the ones where both `Valid` and `Task` are `True`.\nMaking some stats I found that with such conditions just one time series (`0d7ab3a9f9`) in the de dataset have some (88) measures where `StartHesitation == 1`.\nThis makes model selection and cross-validation quite tricky.\nHow are you handling this problem?",
      "votes": null
    },
    {
      "id": "2221933",
      "postDate": "04/14/2023 17:52:35",
      "content": "<p>For now, I train a model for the \"de\" part and another for the \"td\" part.</p>\n<h2>One knows nothing about StartHesitation, I simply ignore it.</h2>\n<p>It might backfire on me later- so I am looking for another solution.</p>",
      "rawMarkdown": "For now, I train a model for the \"de\" part and another for the \"td\" part.\nOne knows nothing about StartHesitation, I simply ignore it.\n- \nIt might backfire on me later- so I am looking for another solution.",
      "votes": null
    },
    {
      "id": "2222049",
      "postDate": "04/14/2023 20:16:43",
      "content": "<p>I see, thank you</p>",
      "rawMarkdown": "I see, thank you",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2221933,
      "author_name": "avivlevi815",
      "author_url": "",
      "post_date": "04/14/2023 17:52:35",
      "content": "<p>For now, I train a model for the \"de\" part and another for the \"td\" part.</p>\n<h2>One knows nothing about StartHesitation, I simply ignore it.</h2>\n<p>It might backfire on me later- so I am looking for another solution.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2222049,
          "author_name": "albertoannoni",
          "author_url": "",
          "post_date": "04/14/2023 20:16:43",
          "content": "<p>I see, thank you</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2220654": "From the competition guide we know that correctly annotated measures from the `de` dataset are the ones where both `Valid` and `Task` are `True`.\nMaking some stats I found that with such conditions just one time series (`0d7ab3a9f9`) in the de dataset have some (88) measures where `StartHesitation == 1`.\nThis makes model selection and cross-validation quite tricky.\nHow are you handling this problem?",
    "2221933": "For now, I train a model for the \"de\" part and another for the \"td\" part.\nOne knows nothing about StartHesitation, I simply ignore it.\n- \nIt might backfire on me later- so I am looking for another solution.",
    "2222049": "I see, thank you"
  },
  "source": "meta"
}