{
  "id": 540388,
  "title": "What is the best result in valid dataset?",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/540388",
  "author_name": "",
  "post_date": "2024-10-14T12:12:21.171185400Z",
  "votes": 5,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I got 0.93 QWK score in my train data and 0.55 in my valid data ? (Obviously, this is due to luck, as many processes are not reproducible.) But i just get 0.463 in LB ,Is it necessary to fine tune this model ? I guess I just over fitting my train dataset.</p>",
  "messages": [
    {
      "id": "3016990",
      "postDate": "10/14/2024 12:12:21",
      "content": "<p>I got 0.93 QWK score in my train data and 0.55 in my valid data ? (Obviously, this is due to luck, as many processes are not reproducible.) But i just get 0.463 in LB ,Is it necessary to fine tune this model ? I guess I just over fitting my train dataset.</p>",
      "rawMarkdown": "I got 0.93 QWK score in my train data and 0.55 in my valid data ? (Obviously, this is due to luck, as many processes are not reproducible.) But i just get 0.463 in LB ,Is it necessary to fine tune this model ? I guess I just over fitting my train dataset.",
      "votes": null
    },
    {
      "id": "3017006",
      "postDate": "10/14/2024 12:29:38",
      "content": "<p>Yes, that’s correct—overfitting. <br>\nConsider using hyperparameters that limit the model's degree of freedom.</p>",
      "rawMarkdown": "Yes, that’s correct—overfitting. \nConsider using hyperparameters that limit the model's degree of freedom.",
      "votes": null
    },
    {
      "id": "3017098",
      "postDate": "10/14/2024 14:15:21",
      "content": "<p>The significant drop in the QWK score from your training (0.93) to validation (0.55) and then to the leaderboard (0.463) suggests overfitting.</p>",
      "rawMarkdown": "The significant drop in the QWK score from your training (0.93) to validation (0.55) and then to the leaderboard (0.463) suggests overfitting.",
      "votes": null
    },
    {
      "id": "3017100",
      "postDate": "10/14/2024 14:16:16",
      "content": "<p>This is 100% overfitting - I also tried the same process as you and got the same kind of result (0.55 CV score). I shall not disclose further but this is obvious <a href=\"https://www.kaggle.com/yashi003\" target=\"_blank\">@yashi003</a> </p>",
      "rawMarkdown": "This is 100% overfitting - I also tried the same process as you and got the same kind of result (0.55 CV score). I shall not disclose further but this is obvious @yashi003",
      "votes": null
    },
    {
      "id": "3017119",
      "postDate": "10/14/2024 14:31:39",
      "content": "<p>Mossad is quite interested in this secrecy. </p>",
      "rawMarkdown": "Mossad is quite interested in this secrecy.",
      "votes": null
    },
    {
      "id": "3017246",
      "postDate": "10/14/2024 17:14:38",
      "content": "<p>Try to filter more features, and will be less overfitting.</p>",
      "rawMarkdown": "Try to filter more features, and will be less overfitting.",
      "votes": null
    },
    {
      "id": "3018487",
      "postDate": "10/15/2024 19:26:43",
      "content": "<p>I got 0.94</p>",
      "rawMarkdown": "I got 0.94",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3017006,
      "author_name": "chanpreetsingh07",
      "author_url": "",
      "post_date": "10/14/2024 12:29:38",
      "content": "<p>Yes, that’s correct—overfitting. <br>\nConsider using hyperparameters that limit the model's degree of freedom.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3017098,
      "author_name": "dirgham",
      "author_url": "",
      "post_date": "10/14/2024 14:15:21",
      "content": "<p>The significant drop in the QWK score from your training (0.93) to validation (0.55) and then to the leaderboard (0.463) suggests overfitting.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3017100,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "10/14/2024 14:16:16",
      "content": "<p>This is 100% overfitting - I also tried the same process as you and got the same kind of result (0.55 CV score). I shall not disclose further but this is obvious <a href=\"https://www.kaggle.com/yashi003\" target=\"_blank\">@yashi003</a> </p>",
      "votes": null,
      "replies": [
        {
          "id": 3017119,
          "author_name": "chanpreetsingh07",
          "author_url": "",
          "post_date": "10/14/2024 14:31:39",
          "content": "<p>Mossad is quite interested in this secrecy. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3017246,
      "author_name": "ggggpeushmy",
      "author_url": "",
      "post_date": "10/14/2024 17:14:38",
      "content": "<p>Try to filter more features, and will be less overfitting.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3018487,
      "author_name": "aimoryou",
      "author_url": "",
      "post_date": "10/15/2024 19:26:43",
      "content": "<p>I got 0.94</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3016990": "I got 0.93 QWK score in my train data and 0.55 in my valid data ? (Obviously, this is due to luck, as many processes are not reproducible.) But i just get 0.463 in LB ,Is it necessary to fine tune this model ? I guess I just over fitting my train dataset.",
    "3017006": "Yes, that’s correct—overfitting. \nConsider using hyperparameters that limit the model's degree of freedom.",
    "3017098": "The significant drop in the QWK score from your training (0.93) to validation (0.55) and then to the leaderboard (0.463) suggests overfitting.",
    "3017100": "This is 100% overfitting - I also tried the same process as you and got the same kind of result (0.55 CV score). I shall not disclose further but this is obvious @yashi003",
    "3017119": "Mossad is quite interested in this secrecy.",
    "3017246": "Try to filter more features, and will be less overfitting.",
    "3018487": "I got 0.94"
  },
  "source": "meta"
}