{
  "id": 593940,
  "title": "Solution with features selected by PLS regression weights",
  "url": "/competitions/drw-crypto-market-prediction/discussion/593940",
  "author_name": "",
  "post_date": "2025-07-31T15:04:38.744082100Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>To attract attention to the solution, I used features selected as most important in PLS regression by weights, 25 best, and then put them to XG Boost. Private score is 0.092, it improved a lot. Unfortunately, I did not check submit to leaderboard. I run it in last minute and was going to run submission for other similar one, but that submission failed. Now I added also engineered features after PLS and private score increased to 0.0976. Notebook XGBoost with features from PLS private score 0.092<br>\n<a href=\"https://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092\" target=\"_blank\">https://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092</a></p>",
  "messages": [
    {
      "id": "3258948",
      "postDate": "07/31/2025 15:04:38",
      "content": "<p>To attract attention to the solution, I used features selected as most important in PLS regression by weights, 25 best, and then put them to XG Boost. Private score is 0.092, it improved a lot. Unfortunately, I did not check submit to leaderboard. I run it in last minute and was going to run submission for other similar one, but that submission failed. Now I added also engineered features after PLS and private score increased to 0.0976. Notebook XGBoost with features from PLS private score 0.092<br>\n<a href=\"https://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092\" target=\"_blank\">https://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092</a></p>",
      "rawMarkdown": "To attract attention to the solution, I used features selected as most important in PLS regression by weights, 25 best, and then put them to XG Boost. Private score is 0.092, it improved a lot. Unfortunately, I did not check submit to leaderboard. I run it in last minute and was going to run submission for other similar one, but that submission failed. Now I added also engineered features after PLS and private score increased to 0.0976. Notebook XGBoost with features from PLS private score 0.092\nhttps://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092",
      "votes": null
    },
    {
      "id": "3259039",
      "postDate": "07/31/2025 18:23:11",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/kuzn137\" target=\"_blank\">@kuzn137</a> ,</p>\n<p>It is super smart to compress features with PLS -- may I ask what the improvement in public/CV score is?</p>",
      "rawMarkdown": "Hi @kuzn137 ,\n\nIt is super smart to compress features with PLS -- may I ask what the improvement in public/CV score is?",
      "votes": null
    },
    {
      "id": "3259110",
      "postDate": "07/31/2025 22:29:44",
      "content": "<p>I did not compressed in this notebook. I used PLS weights as features ranking. I selected here 25 features with best weight. Public score was 0.054. I submitted notebook with compressed features to leaderboard, score increased but a bit, 0.058. </p>",
      "rawMarkdown": "I did not compressed in this notebook. I used PLS weights as features ranking. I selected here 25 features with best weight. Public score was 0.054. I submitted notebook with compressed features to leaderboard, score increased but a bit, 0.058.",
      "votes": null
    },
    {
      "id": "3259113",
      "postDate": "07/31/2025 22:48:02",
      "content": "<p>Not clear for me in your question improvement in respect of which case.  Features will change if model is retrained. Compressed model almost did not changes score when dataset was replaced.</p>",
      "rawMarkdown": "Not clear for me in your question improvement in respect of which case.  Features will change if model is retrained. Compressed model almost did not changes score when dataset was replaced.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3259039,
      "author_name": "alexzhongs",
      "author_url": "",
      "post_date": "07/31/2025 18:23:11",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/kuzn137\" target=\"_blank\">@kuzn137</a> ,</p>\n<p>It is super smart to compress features with PLS -- may I ask what the improvement in public/CV score is?</p>",
      "votes": null,
      "replies": [
        {
          "id": 3259110,
          "author_name": "kuzn137",
          "author_url": "",
          "post_date": "07/31/2025 22:29:44",
          "content": "<p>I did not compressed in this notebook. I used PLS weights as features ranking. I selected here 25 features with best weight. Public score was 0.054. I submitted notebook with compressed features to leaderboard, score increased but a bit, 0.058. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 3259113,
          "author_name": "kuzn137",
          "author_url": "",
          "post_date": "07/31/2025 22:48:02",
          "content": "<p>Not clear for me in your question improvement in respect of which case.  Features will change if model is retrained. Compressed model almost did not changes score when dataset was replaced.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3258948": "To attract attention to the solution, I used features selected as most important in PLS regression by weights, 25 best, and then put them to XG Boost. Private score is 0.092, it improved a lot. Unfortunately, I did not check submit to leaderboard. I run it in last minute and was going to run submission for other similar one, but that submission failed. Now I added also engineered features after PLS and private score increased to 0.0976. Notebook XGBoost with features from PLS private score 0.092\nhttps://www.kaggle.com/code/kuzn137/xgboost-with-features-from-pls-private-score-0-092",
    "3259039": "Hi @kuzn137 ,\n\nIt is super smart to compress features with PLS -- may I ask what the improvement in public/CV score is?",
    "3259110": "I did not compressed in this notebook. I used PLS weights as features ranking. I selected here 25 features with best weight. Public score was 0.054. I submitted notebook with compressed features to leaderboard, score increased but a bit, 0.058.",
    "3259113": "Not clear for me in your question improvement in respect of which case.  Features will change if model is retrained. Compressed model almost did not changes score when dataset was replaced."
  },
  "source": "meta"
}