{
  "id": 317477,
  "title": "How do you tune the hyper parameter after adding a batch of features?",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/317477",
  "author_name": "",
  "post_date": "2022-04-07T09:26:53.386772600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi~~<br>\nDuring this match, I met a problem:<br>\nsometimes the metrics become worse after adding some features, but it's not clear that these features are noisy or the old hyper-parameters become sub-optimal. <br>\nSo my questions are:</p>\n<ol>\n<li>Will you adjust hyper-parameters every time after adding a batch of features?</li>\n<li>If yes,  how to tune them effectively?</li>\n</ol>\n<p>In my case, the ranking model is lightgbm. I usually try some different <code>learning_rate</code>, <code>max_depth</code>, <code>num_leaves</code> and <code>min_child_samples</code> for a better fitting of the new adding features, but it's a little time-consumption. What's your thought??</p>",
  "messages": [
    {
      "id": "1748059",
      "postDate": "04/07/2022 09:26:53",
      "content": "<p>Hi~~<br>\nDuring this match, I met a problem:<br>\nsometimes the metrics become worse after adding some features, but it's not clear that these features are noisy or the old hyper-parameters become sub-optimal. <br>\nSo my questions are:</p>\n<ol>\n<li>Will you adjust hyper-parameters every time after adding a batch of features?</li>\n<li>If yes,  how to tune them effectively?</li>\n</ol>\n<p>In my case, the ranking model is lightgbm. I usually try some different <code>learning_rate</code>, <code>max_depth</code>, <code>num_leaves</code> and <code>min_child_samples</code> for a better fitting of the new adding features, but it's a little time-consumption. What's your thought??</p>",
      "rawMarkdown": "Hi~~\nDuring this match, I met a problem:\nsometimes the metrics become worse after adding some features, but it's not clear that these features are noisy or the old hyper-parameters become sub-optimal. \nSo my questions are:\n1. Will you adjust hyper-parameters every time after adding a batch of features?\n2. If yes,  how to tune them effectively?\n\nIn my case, the ranking model is lightgbm. I usually try some different `learning_rate`, `max_depth`, `num_leaves` and `min_child_samples` for a better fitting of the new adding features, but it's a little time-consumption. What's your thought??",
      "votes": null
    },
    {
      "id": "1748357",
      "postDate": "04/07/2022 13:59:55",
      "content": "<p>Your metrics could perhaps worsen due to added correlations among the features. It is a good practice to revisit the hyperparameters in such cases and fine-tune them after adding one/ more additional features.</p>",
      "rawMarkdown": "Your metrics could perhaps worsen due to added correlations among the features. It is a good practice to revisit the hyperparameters in such cases and fine-tune them after adding one/ more additional features.",
      "votes": null
    },
    {
      "id": "1748861",
      "postDate": "04/08/2022 04:15:55",
      "content": "<p>thanks for your reply~ nice suggestion!</p>",
      "rawMarkdown": "thanks for your reply~ nice suggestion!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1748357,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "04/07/2022 13:59:55",
      "content": "<p>Your metrics could perhaps worsen due to added correlations among the features. It is a good practice to revisit the hyperparameters in such cases and fine-tune them after adding one/ more additional features.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1748861,
          "author_name": "sirius81",
          "author_url": "",
          "post_date": "04/08/2022 04:15:55",
          "content": "<p>thanks for your reply~ nice suggestion!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1748059": "Hi~~\nDuring this match, I met a problem:\nsometimes the metrics become worse after adding some features, but it's not clear that these features are noisy or the old hyper-parameters become sub-optimal. \nSo my questions are:\n1. Will you adjust hyper-parameters every time after adding a batch of features?\n2. If yes,  how to tune them effectively?\n\nIn my case, the ranking model is lightgbm. I usually try some different `learning_rate`, `max_depth`, `num_leaves` and `min_child_samples` for a better fitting of the new adding features, but it's a little time-consumption. What's your thought??",
    "1748357": "Your metrics could perhaps worsen due to added correlations among the features. It is a good practice to revisit the hyperparameters in such cases and fine-tune them after adding one/ more additional features.",
    "1748861": "thanks for your reply~ nice suggestion!"
  },
  "source": "meta"
}