{
  "id": 500293,
  "title": "What the slope 'a' > 0 is meaning ?",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/500293",
  "author_name": "Nakanishi",
  "post_date": "2024-05-05T05:10:43.667000",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I'm curious about why the metrics don't have any penalty for 'a' &gt; 0. It seems possible to achieve high scores in this competition without it. However, I think that 'a' &gt; 0 indicates that the model's Gini score improves as it gets further from the end date of the training dataset. What does this mean? I assumed that without any metric hacking, the distribution of the feature dataset would increasingly resemble that of the training dataset. Does this actually happen in reality?</p>",
  "messages": [
    {
      "id": 2793970,
      "postDate": "2024-05-05T05:10:43.667Z",
      "content": "<p>I'm curious about why the metrics don't have any penalty for 'a' &gt; 0. It seems possible to achieve high scores in this competition without it. However, I think that 'a' &gt; 0 indicates that the model's Gini score improves as it gets further from the end date of the training dataset. What does this mean? I assumed that without any metric hacking, the distribution of the feature dataset would increasingly resemble that of the training dataset. Does this actually happen in reality?</p>",
      "rawMarkdown": "I'm curious about why the metrics don't have any penalty for 'a' > 0. It seems possible to achieve high scores in this competition without it. However, I think that 'a' > 0 indicates that the model's Gini score improves as it gets further from the end date of the training dataset. What does this mean? I assumed that without any metric hacking, the distribution of the feature dataset would increasingly resemble that of the training dataset. Does this actually happen in reality?",
      "votes": 1
    },
    {
      "id": 2794132,
      "postDate": "2024-05-05T06:21:54.300Z",
      "content": "<p>As time goes by, the model's predictions become increasingly accurate (AUC increases)? This is unlikely in reality, unless you intentionally make the model perform poorly at the beginning</p>",
      "rawMarkdown": "As time goes by, the model's predictions become increasingly accurate (AUC increases)? This is unlikely in reality, unless you intentionally make the model perform poorly at the beginning",
      "votes": 2
    },
    {
      "id": 2793991,
      "postDate": "2024-05-05T05:25:43.527Z",
      "content": "<p>Since slope ‘a’ &gt; 0 is unlikely to happen in such a test set, there would be no penalty imposed for this. I won't reveal how I did it until the end of the game, though.</p>",
      "rawMarkdown": "Since slope ‘a’ > 0 is unlikely to happen in such a test set, there would be no penalty imposed for this. I won't reveal how I did it until the end of the game, though."
    },
    {
      "id": 2793990,
      "postDate": "2024-05-05T05:24:46.830Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2794132,
      "author_name": "yunsuxiaozi",
      "author_url": "",
      "post_date": "2024-05-05T06:21:54.300000",
      "content": "<p>As time goes by, the model's predictions become increasingly accurate (AUC increases)? This is unlikely in reality, unless you intentionally make the model perform poorly at the beginning</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 2793991,
      "author_name": "暗黑AGI",
      "author_url": "",
      "post_date": "2024-05-05T05:25:43.527000",
      "content": "<p>Since slope ‘a’ &gt; 0 is unlikely to happen in such a test set, there would be no penalty imposed for this. I won't reveal how I did it until the end of the game, though.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2793990,
      "author_name": "",
      "author_url": "",
      "post_date": "2024-05-05T05:24:46.830000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2793970": "I'm curious about why the metrics don't have any penalty for 'a' > 0. It seems possible to achieve high scores in this competition without it. However, I think that 'a' > 0 indicates that the model's Gini score improves as it gets further from the end date of the training dataset. What does this mean? I assumed that without any metric hacking, the distribution of the feature dataset would increasingly resemble that of the training dataset. Does this actually happen in reality?",
    "2794132": "As time goes by, the model's predictions become increasingly accurate (AUC increases)? This is unlikely in reality, unless you intentionally make the model perform poorly at the beginning",
    "2793991": "Since slope ‘a’ > 0 is unlikely to happen in such a test set, there would be no penalty imposed for this. I won't reveal how I did it until the end of the game, though.",
    "2793990": ""
  }
}