{
  "id": 420158,
  "title": "43th Place Solution",
  "url": "/competitions/predict-student-performance-from-game-play/writeups/konumaru-43th-place-solution",
  "author_name": "",
  "post_date": "2023-07-17T11:10:34.717Z",
  "votes": 13,
  "comment_count": 5,
  "views": 0,
  "content": "<h2>Solution</h2>\n<ul>\n<li>Create features for each level_group.<ul>\n<li>In addition, using the previous level_group features.</li></ul></li>\n<li>LGBM and XGB model for each level.</li>\n<li>Optimize hyperparameters for each level. (Only XGB)</li>\n<li>I think the amount of features is almost the same as what is in the public.</li>\n</ul>\n<h3>Not work for me</h3>\n<ul>\n<li>Catboost model</li>\n<li>level_group probability as feature fo stacking model.</li>\n<li>sample weight for each level.</li>\n<li>optimize threshold of f1-score for each level.</li>\n<li>As a feature of gbdt, using event seqence vectorize with w2v.</li>\n</ul>\n<h3>Not try yet</h3>\n<ul>\n<li>Ensenble knoledge tracing model with transformer or 1dcnn</li>\n<li>Optimize hyperparameters of LGBM for each level.</li>\n</ul>\n<p>repo: <a href=\"https://github.com/konumaru/predict_student_performance\" target=\"_blank\">https://github.com/konumaru/predict_student_performance</a></p>",
  "messages": [
    {
      "id": "2322767",
      "postDate": "06/29/2023 13:38:50",
      "content": "<h2>Solution</h2>\n<ul>\n<li>Create features for each level_group.<ul>\n<li>In addition, using the previous level_group features.</li></ul></li>\n<li>LGBM and XGB model for each level.</li>\n<li>Optimize hyperparameters for each level. (Only XGB)</li>\n<li>I think the amount of features is almost the same as what is in the public.</li>\n</ul>\n<h3>Not work for me</h3>\n<ul>\n<li>Catboost model</li>\n<li>level_group probability as feature fo stacking model.</li>\n<li>sample weight for each level.</li>\n<li>optimize threshold of f1-score for each level.</li>\n<li>As a feature of gbdt, using event seqence vectorize with w2v.</li>\n</ul>\n<h3>Not try yet</h3>\n<ul>\n<li>Ensenble knoledge tracing model with transformer or 1dcnn</li>\n<li>Optimize hyperparameters of LGBM for each level.</li>\n</ul>\n<p>repo: <a href=\"https://github.com/konumaru/predict_student_performance\" target=\"_blank\">https://github.com/konumaru/predict_student_performance</a></p>",
      "rawMarkdown": "## Solution\n\n- Create features for each level_group.\n  - In addition, using the previous level_group features.\n- LGBM and XGB model for each level.\n- Optimize hyperparameters for each level. (Only XGB)\n- I think the amount of features is almost the same as what is in the public.\n\n### Not work for me\n\n- Catboost model\n- level_group probability as feature fo stacking model.\n- sample weight for each level.\n- optimize threshold of f1-score for each level.\n- As a feature of gbdt, using event seqence vectorize with w2v.\n\n### Not try yet\n\n- Ensenble knoledge tracing model with transformer or 1dcnn\n- Optimize hyperparameters of LGBM for each level.\n\n\nrepo: https://github.com/konumaru/predict_student_performance",
      "votes": null
    },
    {
      "id": "2322827",
      "postDate": "06/29/2023 14:14:39",
      "content": "<p>Congratulations for coming 44th <a href=\"https://www.kaggle.com/konumaru\" target=\"_blank\">@konumaru</a> </p>",
      "rawMarkdown": "Congratulations for coming 44th @konumaru",
      "votes": null
    },
    {
      "id": "2323877",
      "postDate": "06/30/2023 08:04:05",
      "content": "<p>Thank you </p>",
      "rawMarkdown": "Thank you",
      "votes": null
    },
    {
      "id": "2324947",
      "postDate": "07/01/2023 03:23:02",
      "content": "<p>Congratulations on your 44th place, well done! <br>\nOne question: Can you tell me some of the results of your experiment? Especially the cv, public score, private score of the previous level group. Thank you!</p>",
      "rawMarkdown": "Congratulations on your 44th place, well done! \nOne question: Can you tell me some of the results of your experiment? Especially the cv, public score, private score of the previous level group. Thank you!",
      "votes": null
    },
    {
      "id": "2325082",
      "postDate": "07/01/2023 05:42:53",
      "content": "<p>Thank you.</p>\n<p>Sorry, but the results of the experiment are not neatly summarized and difficult to share.</p>\n<p>I was looking at the git commit log and SubmissionScore during the experiment, so I will only share the last few submission results so if you have time you can compare them by date, etc.</p>\n<p>commit log: <a href=\"https://github.com/konumaru/predict_student_performance/commits/main\" target=\"_blank\">https://github.com/konumaru/predict_student_performance/commits/main</a><br>\n<code>[update] cv=hogehoge</code></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3460322%2F671e9833bdec7cac9db131573eb142b7%2F2023-07-01%20144143.png?generation=1688190165710104&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Thank you.\n\nSorry, but the results of the experiment are not neatly summarized and difficult to share.\n\nI was looking at the git commit log and SubmissionScore during the experiment, so I will only share the last few submission results so if you have time you can compare them by date, etc.\n\ncommit log: https://github.com/konumaru/predict_student_performance/commits/main\n`[update] cv=hogehoge`\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3460322%2F671e9833bdec7cac9db131573eb142b7%2F2023-07-01%20144143.png?generation=1688190165710104&alt=media)",
      "votes": null
    },
    {
      "id": "2325266",
      "postDate": "07/01/2023 08:04:34",
      "content": "<p>Thank you🎉</p>",
      "rawMarkdown": "Thank you🎉",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2322827,
      "author_name": "swapnilchowdhury",
      "author_url": "",
      "post_date": "06/29/2023 14:14:39",
      "content": "<p>Congratulations for coming 44th <a href=\"https://www.kaggle.com/konumaru\" target=\"_blank\">@konumaru</a> </p>",
      "votes": null,
      "replies": [
        {
          "id": 2323877,
          "author_name": "konumaru",
          "author_url": "",
          "post_date": "06/30/2023 08:04:05",
          "content": "<p>Thank you </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2324947,
      "author_name": "kaggleaau",
      "author_url": "",
      "post_date": "07/01/2023 03:23:02",
      "content": "<p>Congratulations on your 44th place, well done! <br>\nOne question: Can you tell me some of the results of your experiment? Especially the cv, public score, private score of the previous level group. Thank you!</p>",
      "votes": null,
      "replies": [
        {
          "id": 2325082,
          "author_name": "konumaru",
          "author_url": "",
          "post_date": "07/01/2023 05:42:53",
          "content": "<p>Thank you.</p>\n<p>Sorry, but the results of the experiment are not neatly summarized and difficult to share.</p>\n<p>I was looking at the git commit log and SubmissionScore during the experiment, so I will only share the last few submission results so if you have time you can compare them by date, etc.</p>\n<p>commit log: <a href=\"https://github.com/konumaru/predict_student_performance/commits/main\" target=\"_blank\">https://github.com/konumaru/predict_student_performance/commits/main</a><br>\n<code>[update] cv=hogehoge</code></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3460322%2F671e9833bdec7cac9db131573eb142b7%2F2023-07-01%20144143.png?generation=1688190165710104&amp;alt=media\" alt=\"\"></p>",
          "votes": null,
          "replies": [
            {
              "id": 2325266,
              "author_name": "kaggleaau",
              "author_url": "",
              "post_date": "07/01/2023 08:04:34",
              "content": "<p>Thank you🎉</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2322767": "## Solution\n\n- Create features for each level_group.\n  - In addition, using the previous level_group features.\n- LGBM and XGB model for each level.\n- Optimize hyperparameters for each level. (Only XGB)\n- I think the amount of features is almost the same as what is in the public.\n\n### Not work for me\n\n- Catboost model\n- level_group probability as feature fo stacking model.\n- sample weight for each level.\n- optimize threshold of f1-score for each level.\n- As a feature of gbdt, using event seqence vectorize with w2v.\n\n### Not try yet\n\n- Ensenble knoledge tracing model with transformer or 1dcnn\n- Optimize hyperparameters of LGBM for each level.\n\n\nrepo: https://github.com/konumaru/predict_student_performance",
    "2322827": "Congratulations for coming 44th @konumaru",
    "2323877": "Thank you",
    "2324947": "Congratulations on your 44th place, well done! \nOne question: Can you tell me some of the results of your experiment? Especially the cv, public score, private score of the previous level group. Thank you!",
    "2325082": "Thank you.\n\nSorry, but the results of the experiment are not neatly summarized and difficult to share.\n\nI was looking at the git commit log and SubmissionScore during the experiment, so I will only share the last few submission results so if you have time you can compare them by date, etc.\n\ncommit log: https://github.com/konumaru/predict_student_performance/commits/main\n`[update] cv=hogehoge`\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3460322%2F671e9833bdec7cac9db131573eb142b7%2F2023-07-01%20144143.png?generation=1688190165710104&alt=media)",
    "2325266": "Thank you🎉"
  },
  "source": "meta"
}