{
  "id": 507539,
  "title": "Meta Featuring for this competition",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/507539",
  "author_name": "Andreas Bisiadis",
  "post_date": "2024-05-26T08:25:45.114000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>For the majority of Kaggle competitions, the gold-medal solutions include meta featuring - using OOF predictions as a feature column. Thus, it's logical to believe that the top scoring solutions of this competition (not the ones with metric hacking!) have included OOF predictions for the models. </p>\n<p>Since there are almost no notebooks explaining the rationale behind meta-featuring, has anyone implemented this approach? if yes, how did OOF predictions boosted the CV/LB scores?</p>",
  "messages": [
    {
      "id": 2837024,
      "postDate": "2024-05-26T08:25:45.113Z",
      "content": "<p>For the majority of Kaggle competitions, the gold-medal solutions include meta featuring - using OOF predictions as a feature column. Thus, it's logical to believe that the top scoring solutions of this competition (not the ones with metric hacking!) have included OOF predictions for the models. </p>\n<p>Since there are almost no notebooks explaining the rationale behind meta-featuring, has anyone implemented this approach? if yes, how did OOF predictions boosted the CV/LB scores?</p>",
      "rawMarkdown": "For the majority of Kaggle competitions, the gold-medal solutions include meta featuring - using OOF predictions as a feature column. Thus, it's logical to believe that the top scoring solutions of this competition (not the ones with metric hacking!) have included OOF predictions for the models. \n\nSince there are almost no notebooks explaining the rationale behind meta-featuring, has anyone implemented this approach? if yes, how did OOF predictions boosted the CV/LB scores?",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2837024": "For the majority of Kaggle competitions, the gold-medal solutions include meta featuring - using OOF predictions as a feature column. Thus, it's logical to believe that the top scoring solutions of this competition (not the ones with metric hacking!) have included OOF predictions for the models. \n\nSince there are almost no notebooks explaining the rationale behind meta-featuring, has anyone implemented this approach? if yes, how did OOF predictions boosted the CV/LB scores?"
  }
}