{
  "id": 584453,
  "title": "Should I Optimize Hyperparameters?",
  "url": "/competitions/drw-crypto-market-prediction/discussion/584453",
  "author_name": "",
  "post_date": "2025-06-13T13:10:52.723347100Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi there, I am a Kaggle Novice and this is my first time attending a competition. Firstly, thanks a lot for the generous sharing, I have learnt a lot from them.</p>\n<p>I noticed that most LGBM models in the public codes use the same hyperparameters. Those hyperparas seem like optimized because I have ever tried to tune a bit and get a worse score. The notebook <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/faressayah/hyperparameter-optimization-for-machine-learning</a> introduced the optimization methods. My question (may be naive) is:</p>\n<ul>\n<li>Should I spend time on tuning the hyperparas or stay focused on mining features currently?</li>\n</ul>\n<p>The following questions are:</p>\n<ul>\n<li>If I add a new feature to the model, should I re-optimize the hyperparas?</li>\n<li>Will the optimized hyperparas lead to the over-fitting problem with the training data/public score?</li>\n</ul>\n<p>Appreciate your kind reply in advance!</p>",
  "messages": [
    {
      "id": "3223538",
      "postDate": "06/13/2025 13:10:52",
      "content": "<p>Hi there, I am a Kaggle Novice and this is my first time attending a competition. Firstly, thanks a lot for the generous sharing, I have learnt a lot from them.</p>\n<p>I noticed that most LGBM models in the public codes use the same hyperparameters. Those hyperparas seem like optimized because I have ever tried to tune a bit and get a worse score. The notebook <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/faressayah/hyperparameter-optimization-for-machine-learning</a> introduced the optimization methods. My question (may be naive) is:</p>\n<ul>\n<li>Should I spend time on tuning the hyperparas or stay focused on mining features currently?</li>\n</ul>\n<p>The following questions are:</p>\n<ul>\n<li>If I add a new feature to the model, should I re-optimize the hyperparas?</li>\n<li>Will the optimized hyperparas lead to the over-fitting problem with the training data/public score?</li>\n</ul>\n<p>Appreciate your kind reply in advance!</p>",
      "rawMarkdown": "Hi there, I am a Kaggle Novice and this is my first time attending a competition. Firstly, thanks a lot for the generous sharing, I have learnt a lot from them.\n\nI noticed that most LGBM models in the public codes use the same hyperparameters. Those hyperparas seem like optimized because I have ever tried to tune a bit and get a worse score. The notebook [https://www.kaggle.com/code/faressayah/hyperparameter-optimization-for-machine-learning](url) introduced the optimization methods. My question (may be naive) is:\n- Should I spend time on tuning the hyperparas or stay focused on mining features currently?\n\nThe following questions are:\n- If I add a new feature to the model, should I re-optimize the hyperparas?\n- Will the optimized hyperparas lead to the over-fitting problem with the training data/public score?\n\nAppreciate your kind reply in advance!",
      "votes": null
    },
    {
      "id": "3224004",
      "postDate": "06/14/2025 08:00:47",
      "content": "<p>Do not try to tune it in the early stage of the competition.</p>",
      "rawMarkdown": "Do not try to tune it in the early stage of the competition.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3224004,
      "author_name": "yutongzhang20080108",
      "author_url": "",
      "post_date": "06/14/2025 08:00:47",
      "content": "<p>Do not try to tune it in the early stage of the competition.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3223538": "Hi there, I am a Kaggle Novice and this is my first time attending a competition. Firstly, thanks a lot for the generous sharing, I have learnt a lot from them.\n\nI noticed that most LGBM models in the public codes use the same hyperparameters. Those hyperparas seem like optimized because I have ever tried to tune a bit and get a worse score. The notebook [https://www.kaggle.com/code/faressayah/hyperparameter-optimization-for-machine-learning](url) introduced the optimization methods. My question (may be naive) is:\n- Should I spend time on tuning the hyperparas or stay focused on mining features currently?\n\nThe following questions are:\n- If I add a new feature to the model, should I re-optimize the hyperparas?\n- Will the optimized hyperparas lead to the over-fitting problem with the training data/public score?\n\nAppreciate your kind reply in advance!",
    "3224004": "Do not try to tune it in the early stage of the competition."
  },
  "source": "meta"
}