{
  "id": 482941,
  "title": "Whats going on? [i know im late]",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/482941",
  "author_name": "TB Dukale",
  "post_date": "2024-03-10T06:47:52.397000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Can someone bring me up to speed on what happened and why we need to alter some of the datasets?</p>\n<ol>\n<li>I have read the latest comment of the host abt altering \"WEEK_NUM\" while keeping the evaluation metrics. [the way i understood it] The evaluation metrics uses some weekly aggregations before applying the eval metrics. Which clearly means somewhere &amp; somewhat we need to \"feature engineer\" &amp; create a new feature related to the \"WEEK_NUM\". </li>\n<li>Will they alter the \"date_decision\" &amp; \"month\"  columns too? cause it will have no meaning dropping the WEEK_NUM and leaving these two up there. </li>\n<li>isn't it wise to keep the dataset as it is and change the eval metrics/way of evaluation to some thing like \"replacing the test dataset to a new one with week_num &gt; 91\"</li>\n<li>I work for a financial institution. I know that our internal eval metrics differs from time to time for many reasons. [the available money we have/ peace and stability of our country/ gov policy etc]. a person who is eligible today might not be eligible after 6 months.  how am i supposed to infer those things to the model without \"date_decision\" or \"week_num\"?</li>\n<li>lastly, What did my \"colleagues\" did to create all this turmoil?</li>\n</ol>\n<p>Thank You.</p>\n<p>Edited: Note that there are multiple datetime columns in the competition dataset from which one can infer and approximates \"date_decision\".  What are we gona do abt them?</p>",
  "messages": [
    {
      "id": 2689855,
      "postDate": "2024-03-10T06:47:52.397Z",
      "content": "<p>Can someone bring me up to speed on what happened and why we need to alter some of the datasets?</p>\n<ol>\n<li>I have read the latest comment of the host abt altering \"WEEK_NUM\" while keeping the evaluation metrics. [the way i understood it] The evaluation metrics uses some weekly aggregations before applying the eval metrics. Which clearly means somewhere &amp; somewhat we need to \"feature engineer\" &amp; create a new feature related to the \"WEEK_NUM\". </li>\n<li>Will they alter the \"date_decision\" &amp; \"month\"  columns too? cause it will have no meaning dropping the WEEK_NUM and leaving these two up there. </li>\n<li>isn't it wise to keep the dataset as it is and change the eval metrics/way of evaluation to some thing like \"replacing the test dataset to a new one with week_num &gt; 91\"</li>\n<li>I work for a financial institution. I know that our internal eval metrics differs from time to time for many reasons. [the available money we have/ peace and stability of our country/ gov policy etc]. a person who is eligible today might not be eligible after 6 months.  how am i supposed to infer those things to the model without \"date_decision\" or \"week_num\"?</li>\n<li>lastly, What did my \"colleagues\" did to create all this turmoil?</li>\n</ol>\n<p>Thank You.</p>\n<p>Edited: Note that there are multiple datetime columns in the competition dataset from which one can infer and approximates \"date_decision\".  What are we gona do abt them?</p>",
      "rawMarkdown": "Can someone bring me up to speed on what happened and why we need to alter some of the datasets?\n1. I have read the latest comment of the host abt altering \"WEEK_NUM\" while keeping the evaluation metrics. [the way i understood it] The evaluation metrics uses some weekly aggregations before applying the eval metrics. Which clearly means somewhere & somewhat we need to \"feature engineer\" & create a new feature related to the \"WEEK_NUM\". \n2. Will they alter the \"date_decision\" & \"month\"  columns too? cause it will have no meaning dropping the WEEK_NUM and leaving these two up there. \n3. isn't it wise to keep the dataset as it is and change the eval metrics/way of evaluation to some thing like \"replacing the test dataset to a new one with week_num > 91\"\n4. I work for a financial institution. I know that our internal eval metrics differs from time to time for many reasons. [the available money we have/ peace and stability of our country/ gov policy etc]. a person who is eligible today might not be eligible after 6 months.  how am i supposed to infer those things to the model without \"date_decision\" or \"week_num\"?\n5. lastly, What did my \"colleagues\" did to create all this turmoil?\n\nThank You.\n\nEdited: Note that there are multiple datetime columns in the competition dataset from which one can infer and approximates \"date_decision\".  What are we gona do abt them?"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2689855": "Can someone bring me up to speed on what happened and why we need to alter some of the datasets?\n1. I have read the latest comment of the host abt altering \"WEEK_NUM\" while keeping the evaluation metrics. [the way i understood it] The evaluation metrics uses some weekly aggregations before applying the eval metrics. Which clearly means somewhere & somewhat we need to \"feature engineer\" & create a new feature related to the \"WEEK_NUM\". \n2. Will they alter the \"date_decision\" & \"month\"  columns too? cause it will have no meaning dropping the WEEK_NUM and leaving these two up there. \n3. isn't it wise to keep the dataset as it is and change the eval metrics/way of evaluation to some thing like \"replacing the test dataset to a new one with week_num > 91\"\n4. I work for a financial institution. I know that our internal eval metrics differs from time to time for many reasons. [the available money we have/ peace and stability of our country/ gov policy etc]. a person who is eligible today might not be eligible after 6 months.  how am i supposed to infer those things to the model without \"date_decision\" or \"week_num\"?\n5. lastly, What did my \"colleagues\" did to create all this turmoil?\n\nThank You.\n\nEdited: Note that there are multiple datetime columns in the competition dataset from which one can infer and approximates \"date_decision\".  What are we gona do abt them?"
  }
}