{
  "id": 508027,
  "title": "Hack by 'max_pmts_year_1139T' to 0.617",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/508027",
  "author_name": "Jack Lee",
  "post_date": "2024-05-28T04:59:25.238000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>It is worth mentioning that the maximum value of the aggregated pmts_year_1139T is likely close to the actual year. In the training set, this value ranges from 2018 to 2021 (with only a few in 2018), and it has a significant correlation with WEEK_NUM. By adjusting the scores of the earlier years, we can perform metric hacking. By filtering for values less than or equal to the median, we achieved a score of 0.617. Unfortunately, when we ultimately chose to filter by values less than the median, the results were not satisfactory.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14376777%2F691affcafba91a1627ea8e6cc4e6d378%2F_20240528125321.png?generation=1716872021066170&amp;alt=media\" alt=\"score\"></p>",
  "messages": [
    {
      "id": 2840403,
      "postDate": "2024-05-28T04:59:25.237Z",
      "content": "<p>It is worth mentioning that the maximum value of the aggregated pmts_year_1139T is likely close to the actual year. In the training set, this value ranges from 2018 to 2021 (with only a few in 2018), and it has a significant correlation with WEEK_NUM. By adjusting the scores of the earlier years, we can perform metric hacking. By filtering for values less than or equal to the median, we achieved a score of 0.617. Unfortunately, when we ultimately chose to filter by values less than the median, the results were not satisfactory.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14376777%2F691affcafba91a1627ea8e6cc4e6d378%2F_20240528125321.png?generation=1716872021066170&amp;alt=media\" alt=\"score\"></p>",
      "rawMarkdown": "It is worth mentioning that the maximum value of the aggregated pmts_year_1139T is likely close to the actual year. In the training set, this value ranges from 2018 to 2021 (with only a few in 2018), and it has a significant correlation with WEEK_NUM. By adjusting the scores of the earlier years, we can perform metric hacking. By filtering for values less than or equal to the median, we achieved a score of 0.617. Unfortunately, when we ultimately chose to filter by values less than the median, the results were not satisfactory.\n\n![score](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14376777%2F691affcafba91a1627ea8e6cc4e6d378%2F_20240528125321.png?generation=1716872021066170&alt=media)",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2840403": "It is worth mentioning that the maximum value of the aggregated pmts_year_1139T is likely close to the actual year. In the training set, this value ranges from 2018 to 2021 (with only a few in 2018), and it has a significant correlation with WEEK_NUM. By adjusting the scores of the earlier years, we can perform metric hacking. By filtering for values less than or equal to the median, we achieved a score of 0.617. Unfortunately, when we ultimately chose to filter by values less than the median, the results were not satisfactory.\n\n![score](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14376777%2F691affcafba91a1627ea8e6cc4e6d378%2F_20240528125321.png?generation=1716872021066170&alt=media)"
  }
}