{
  "id": 552487,
  "title": "How to survive the volatility？I did part of that",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/552487",
  "author_name": "",
  "post_date": "2024-12-20T00:52:49.983038600Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>As expected, another huge shake up. You are welcome to share my discussion on how to survive big fluctuations. I will share my experience in a few hours. First, I have to get some sleep🥱</p>\n<p>Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it<br>\n<a href=\"https://www.kaggle.com/code/ruichardliu/baseliner0-445\" target=\"_blank\">https://www.kaggle.com/code/ruichardliu/baseliner0-445</a></p>",
  "messages": [
    {
      "id": "3076426",
      "postDate": "12/20/2024 00:52:49",
      "content": "<p>As expected, another huge shake up. You are welcome to share my discussion on how to survive big fluctuations. I will share my experience in a few hours. First, I have to get some sleep🥱</p>\n<p>Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it<br>\n<a href=\"https://www.kaggle.com/code/ruichardliu/baseliner0-445\" target=\"_blank\">https://www.kaggle.com/code/ruichardliu/baseliner0-445</a></p>",
      "rawMarkdown": "As expected, another huge shake up. You are welcome to share my discussion on how to survive big fluctuations. I will share my experience in a few hours. First, I have to get some sleep🥱\n\nHere is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it\nhttps://www.kaggle.com/code/ruichardliu/baseliner0-445",
      "votes": null
    },
    {
      "id": "3076433",
      "postDate": "12/20/2024 00:56:21",
      "content": "<ol>\n<li>Do not use actigraphy features</li>\n<li>Only use a subset of features from the main data. I only used ['CGAS-CGAS_Score', 'PreInt_EduHx-computerinternet_hoursday', \"Basic_Demos-Age\", \"SDS-SDS_Total_Raw\", \"SDS-SDS_Total_T\", \"Basic_Demos-Sex\"] in my models (0.421 -&gt; 0.438)</li>\n<li>Manually adjust the threshold by a little bit (don’t trust tuned threshold to the CV).  (0.438 -&gt; 0.452) </li>\n<li>Impute the missing values with KNNImputer (0.452 -&gt; 0.466)</li>\n</ol>\n<p>With this I was able to get Private LB 0.466. But sadly I didn’t do (3) and (4) cos I felt they were questionable and worsened the CV, and got 0.438. My public and private scores were quite close </p>\n<p><a href=\"https://www.kaggle.com/code/yeoyunsianggeremie/pb-0-466-single-multiseed-catboost-with-6-feats?scriptVersionId=213940183\" target=\"_blank\">Notebook</a> - the thresholds were reduced by 0.05 here</p>\n<p>Idea here is to reduce dataset noise and only keep features that are relevant both in the model and domain perspective</p>",
      "rawMarkdown": "1. Do not use actigraphy features\n2. Only use a subset of features from the main data. I only used ['CGAS-CGAS_Score', 'PreInt_EduHx-computerinternet_hoursday', \"Basic_Demos-Age\", \"SDS-SDS_Total_Raw\", \"SDS-SDS_Total_T\", \"Basic_Demos-Sex\"] in my models (0.421 -> 0.438)\n3. Manually adjust the threshold by a little bit (don’t trust tuned threshold to the CV).  (0.438 -> 0.452) \n4. Impute the missing values with KNNImputer (0.452 -> 0.466)\n\nWith this I was able to get Private LB 0.466. But sadly I didn’t do (3) and (4) cos I felt they were questionable and worsened the CV, and got 0.438. My public and private scores were quite close \n\n[Notebook](https://www.kaggle.com/code/yeoyunsianggeremie/pb-0-466-single-multiseed-catboost-with-6-feats?scriptVersionId=213940183) - the thresholds were reduced by 0.05 here\n\nIdea here is to reduce dataset noise and only keep features that are relevant both in the model and domain perspective",
      "votes": null
    },
    {
      "id": "3076448",
      "postDate": "12/20/2024 01:06:50",
      "content": "<p><code>Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it</code></p>\n<p>I think it is still private <a href=\"https://www.kaggle.com/ruichardliu\" target=\"_blank\">@ruichardliu</a> </p>",
      "rawMarkdown": "`Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it`\n\nI think it is still private @ruichardliu",
      "votes": null
    },
    {
      "id": "3076454",
      "postDate": "12/20/2024 01:10:02",
      "content": "<p><a href=\"https://www.kaggle.com/code/ruichardliu/baseliner0-445\" target=\"_blank\">https://www.kaggle.com/code/ruichardliu/baseliner0-445</a></p>\n<p>I forgot to open it. I'm sorry. It should be okay now</p>",
      "rawMarkdown": "https://www.kaggle.com/code/ruichardliu/baseliner0-445\n\nI forgot to open it. I'm sorry. It should be okay now",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3076433,
      "author_name": "yeoyunsianggeremie",
      "author_url": "",
      "post_date": "12/20/2024 00:56:21",
      "content": "<ol>\n<li>Do not use actigraphy features</li>\n<li>Only use a subset of features from the main data. I only used ['CGAS-CGAS_Score', 'PreInt_EduHx-computerinternet_hoursday', \"Basic_Demos-Age\", \"SDS-SDS_Total_Raw\", \"SDS-SDS_Total_T\", \"Basic_Demos-Sex\"] in my models (0.421 -&gt; 0.438)</li>\n<li>Manually adjust the threshold by a little bit (don’t trust tuned threshold to the CV).  (0.438 -&gt; 0.452) </li>\n<li>Impute the missing values with KNNImputer (0.452 -&gt; 0.466)</li>\n</ol>\n<p>With this I was able to get Private LB 0.466. But sadly I didn’t do (3) and (4) cos I felt they were questionable and worsened the CV, and got 0.438. My public and private scores were quite close </p>\n<p><a href=\"https://www.kaggle.com/code/yeoyunsianggeremie/pb-0-466-single-multiseed-catboost-with-6-feats?scriptVersionId=213940183\" target=\"_blank\">Notebook</a> - the thresholds were reduced by 0.05 here</p>\n<p>Idea here is to reduce dataset noise and only keep features that are relevant both in the model and domain perspective</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3076448,
      "author_name": "abdmental01",
      "author_url": "",
      "post_date": "12/20/2024 01:06:50",
      "content": "<p><code>Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it</code></p>\n<p>I think it is still private <a href=\"https://www.kaggle.com/ruichardliu\" target=\"_blank\">@ruichardliu</a> </p>",
      "votes": null,
      "replies": [
        {
          "id": 3076454,
          "author_name": "ruichardliu",
          "author_url": "",
          "post_date": "12/20/2024 01:10:02",
          "content": "<p><a href=\"https://www.kaggle.com/code/ruichardliu/baseliner0-445\" target=\"_blank\">https://www.kaggle.com/code/ruichardliu/baseliner0-445</a></p>\n<p>I forgot to open it. I'm sorry. It should be okay now</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3076426": "As expected, another huge shake up. You are welcome to share my discussion on how to survive big fluctuations. I will share my experience in a few hours. First, I have to get some sleep🥱\n\nHere is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it\nhttps://www.kaggle.com/code/ruichardliu/baseliner0-445",
    "3076433": "1. Do not use actigraphy features\n2. Only use a subset of features from the main data. I only used ['CGAS-CGAS_Score', 'PreInt_EduHx-computerinternet_hoursday', \"Basic_Demos-Age\", \"SDS-SDS_Total_Raw\", \"SDS-SDS_Total_T\", \"Basic_Demos-Sex\"] in my models (0.421 -> 0.438)\n3. Manually adjust the threshold by a little bit (don’t trust tuned threshold to the CV).  (0.438 -> 0.452) \n4. Impute the missing values with KNNImputer (0.452 -> 0.466)\n\nWith this I was able to get Private LB 0.466. But sadly I didn’t do (3) and (4) cos I felt they were questionable and worsened the CV, and got 0.438. My public and private scores were quite close \n\n[Notebook](https://www.kaggle.com/code/yeoyunsianggeremie/pb-0-466-single-multiseed-catboost-with-6-feats?scriptVersionId=213940183) - the thresholds were reduced by 0.05 here\n\nIdea here is to reduce dataset noise and only keep features that are relevant both in the model and domain perspective",
    "3076448": "`Here is my LB:0.440 PB:0.477 notebook, unfortunately I did not choose it`\n\nI think it is still private @ruichardliu",
    "3076454": "https://www.kaggle.com/code/ruichardliu/baseliner0-445\n\nI forgot to open it. I'm sorry. It should be okay now"
  },
  "source": "meta"
}