{
  "id": 189074,
  "title": "More features",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/189074",
  "author_name": "Alexey Poddiachyi",
  "post_date": "2020-10-06T15:02:42.677000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Why do every time I add features from another notebook I get some ridiculus score like -24.5929?<br>\nIs there some hidden rule I don't know about?</p>\n<p>Meanwhile, my CV score is -6.7217.</p>",
  "messages": [
    {
      "id": 1039678,
      "postDate": "2020-10-06T17:59:32.763Z",
      "content": "<p>Hi! Hard to say without specifics, but I suspect it might be because of the certainty field. The modified Laplace log likelihood loss function penalises severely when the wrong FVC prediction is associated with a high certainty (ie, a low number in the \"Certainty\" field). Therefore, at a conceptual level, these very low scores could indicate that your new predictions incorporating the features from other notebooks are wronger for the same level of certainty, or that they are producing very certain predictions that are ultimately wrong.</p>",
      "rawMarkdown": "Hi! Hard to say without specifics, but I suspect it might be because of the certainty field. The modified Laplace log likelihood loss function penalises severely when the wrong FVC prediction is associated with a high certainty (ie, a low number in the \"Certainty\" field). Therefore, at a conceptual level, these very low scores could indicate that your new predictions incorporating the features from other notebooks are wronger for the same level of certainty, or that they are producing very certain predictions that are ultimately wrong."
    },
    {
      "id": 1039411,
      "postDate": "2020-10-06T15:02:42.677Z",
      "content": "<p>Why do every time I add features from another notebook I get some ridiculus score like -24.5929?<br>\nIs there some hidden rule I don't know about?</p>\n<p>Meanwhile, my CV score is -6.7217.</p>",
      "rawMarkdown": "Why do every time I add features from another notebook I get some ridiculus score like -24.5929?\nIs there some hidden rule I don't know about?\n\nMeanwhile, my CV score is -6.7217."
    },
    {
      "id": 1039423,
      "postDate": "2020-10-06T15:11:23.483Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1039678,
      "author_name": "Douglas K.G. Araujo",
      "author_url": "",
      "post_date": "2020-10-06T17:59:32.763000",
      "content": "<p>Hi! Hard to say without specifics, but I suspect it might be because of the certainty field. The modified Laplace log likelihood loss function penalises severely when the wrong FVC prediction is associated with a high certainty (ie, a low number in the \"Certainty\" field). Therefore, at a conceptual level, these very low scores could indicate that your new predictions incorporating the features from other notebooks are wronger for the same level of certainty, or that they are producing very certain predictions that are ultimately wrong.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1039423,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-10-06T15:11:23.483000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1039678": "Hi! Hard to say without specifics, but I suspect it might be because of the certainty field. The modified Laplace log likelihood loss function penalises severely when the wrong FVC prediction is associated with a high certainty (ie, a low number in the \"Certainty\" field). Therefore, at a conceptual level, these very low scores could indicate that your new predictions incorporating the features from other notebooks are wronger for the same level of certainty, or that they are producing very certain predictions that are ultimately wrong.",
    "1039411": "Why do every time I add features from another notebook I get some ridiculus score like -24.5929?\nIs there some hidden rule I don't know about?\n\nMeanwhile, my CV score is -6.7217.",
    "1039423": ""
  }
}