{
  "id": 416053,
  "title": "Combine or not combine defog with tdcsfog",
  "url": "/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/416053",
  "author_name": "Ernest Glukhov",
  "post_date": "2023-06-09T10:41:25.626000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I would be very interested to know if you used a strategy of combining two datasets or trained separate models?<br>\nWe tried to train separate models, as the results on two datasets were very different. And the metric performed well on the public score.<br>\nHowever, we have dropped very much on the private dataset: 0.404-&gt;0.23.<br>\nI suspect that the division into defog and tdcsfog in the private dataset was different, which is quite a shame.</p>",
  "messages": [
    {
      "id": 2293620,
      "postDate": "2023-06-09T10:41:25.627Z",
      "content": "<p>I would be very interested to know if you used a strategy of combining two datasets or trained separate models?<br>\nWe tried to train separate models, as the results on two datasets were very different. And the metric performed well on the public score.<br>\nHowever, we have dropped very much on the private dataset: 0.404-&gt;0.23.<br>\nI suspect that the division into defog and tdcsfog in the private dataset was different, which is quite a shame.</p>",
      "rawMarkdown": "I would be very interested to know if you used a strategy of combining two datasets or trained separate models?\nWe tried to train separate models, as the results on two datasets were very different. And the metric performed well on the public score.\nHowever, we have dropped very much on the private dataset: 0.404->0.23.\nI suspect that the division into defog and tdcsfog in the private dataset was different, which is quite a shame.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2293620": "I would be very interested to know if you used a strategy of combining two datasets or trained separate models?\nWe tried to train separate models, as the results on two datasets were very different. And the metric performed well on the public score.\nHowever, we have dropped very much on the private dataset: 0.404->0.23.\nI suspect that the division into defog and tdcsfog in the private dataset was different, which is quite a shame."
  }
}