{
  "id": 501553,
  "title": "About basic code revise",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/501553",
  "author_name": "",
  "post_date": "2024-05-09T17:23:03.851791900Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>So, as a new beginner, what I only know is to revise the highest public score's basic code template, but I really don't know which part to focus on. Should I focus on data cleaning, feature selection, parameter adjustment or model votting or something else. Really don't get which part(s) is done quite well at present as to focus on the real trouble to possibly improve the score. Thanks.</p>",
  "messages": [
    {
      "id": "2803888",
      "postDate": "05/09/2024 17:23:03",
      "content": "<p>So, as a new beginner, what I only know is to revise the highest public score's basic code template, but I really don't know which part to focus on. Should I focus on data cleaning, feature selection, parameter adjustment or model votting or something else. Really don't get which part(s) is done quite well at present as to focus on the real trouble to possibly improve the score. Thanks.</p>",
      "rawMarkdown": "So, as a new beginner, what I only know is to revise the highest public score's basic code template, but I really don't know which part to focus on. Should I focus on data cleaning, feature selection, parameter adjustment or model votting or something else. Really don't get which part(s) is done quite well at present as to focus on the real trouble to possibly improve the score. Thanks.",
      "votes": null
    },
    {
      "id": "2805015",
      "postDate": "05/10/2024 10:13:41",
      "content": "<p>To improve the score now everyone is concentrated on the metric hacking.<br>\nBut as for a beginner - just take your pace to learn as much as possible and don't concentrate at the score. As soon as you get &gt;0.580 you are not bad already.<br>\nThe public code template is used in almost every public notebook with small differences so you can play with aggregations methods. As for the modeling I didn't saw anything special in public - the splitting based on WEEK_NUM and normal training. Here is an implementation of the competition metric directly into the training process <a href=\"https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577\" target=\"_blank\">https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577</a></p>",
      "rawMarkdown": "To improve the score now everyone is concentrated on the metric hacking.\nBut as for a beginner - just take your pace to learn as much as possible and don't concentrate at the score. As soon as you get >0.580 you are not bad already.\nThe public code template is used in almost every public notebook with small differences so you can play with aggregations methods. As for the modeling I didn't saw anything special in public - the splitting based on WEEK_NUM and normal training. Here is an implementation of the competition metric directly into the training process https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577",
      "votes": null
    },
    {
      "id": "2809284",
      "postDate": "05/12/2024 16:41:55",
      "content": "<p>Thanks for advice! I would consider your recommendation!</p>",
      "rawMarkdown": "Thanks for advice! I would consider your recommendation!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2805015,
      "author_name": "eu1234",
      "author_url": "",
      "post_date": "05/10/2024 10:13:41",
      "content": "<p>To improve the score now everyone is concentrated on the metric hacking.<br>\nBut as for a beginner - just take your pace to learn as much as possible and don't concentrate at the score. As soon as you get &gt;0.580 you are not bad already.<br>\nThe public code template is used in almost every public notebook with small differences so you can play with aggregations methods. As for the modeling I didn't saw anything special in public - the splitting based on WEEK_NUM and normal training. Here is an implementation of the competition metric directly into the training process <a href=\"https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577\" target=\"_blank\">https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 2809284,
          "author_name": "xinyc39",
          "author_url": "",
          "post_date": "05/12/2024 16:41:55",
          "content": "<p>Thanks for advice! I would consider your recommendation!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2803888": "So, as a new beginner, what I only know is to revise the highest public score's basic code template, but I really don't know which part to focus on. Should I focus on data cleaning, feature selection, parameter adjustment or model votting or something else. Really don't get which part(s) is done quite well at present as to focus on the real trouble to possibly improve the score. Thanks.",
    "2805015": "To improve the score now everyone is concentrated on the metric hacking.\nBut as for a beginner - just take your pace to learn as much as possible and don't concentrate at the score. As soon as you get >0.580 you are not bad already.\nThe public code template is used in almost every public notebook with small differences so you can play with aggregations methods. As for the modeling I didn't saw anything special in public - the splitting based on WEEK_NUM and normal training. Here is an implementation of the competition metric directly into the training process https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/501577",
    "2809284": "Thanks for advice! I would consider your recommendation!"
  },
  "source": "meta"
}