{
  "id": 552759,
  "title": "【EN/JP】Solution - DeepAutoEncorder & QWK Optimize - Private Score@0.445　20241226update",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/552759",
  "author_name": "fuku4ki",
  "post_date": "2024-12-21T11:08:13.302000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>---@English<br>\nHere is the solution I would like to share.</p>\n<p>Since there was a significant shakeup on the Public Leaderboard, I don't believe this is a groundbreaking solution.</p>\n<p>The main approaches implemented are as follows:</p>\n<p>Aggregating and using DeepAutoEncoder for TimeSeries data<br>\nFeature Engineering<br>\nRemoving rows without the target variable<br>\nOptimizing the three boundary values for the evaluation metric using Optuna<br>\nApplying Seed Averaging (although the single model performed better)<br>\nUsing LightGBM, while also experimenting with XGBoost, CatBoost, and TabNet.</p>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder</a></p>\n<p>Addition:<br>\nThe attached file is a script that was executed on Google Colab. The seed values were manually switched, and a Seed Average was applied for the final submission.<br>\nFor the final submission, we based it on the model and thresholds obtained from Google Colab, and used a majority vote across different seeds for submission.</p>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission</a></p>\n<p>----@JP<br>\nソリューションを共有します。</p>\n<p>Public LeaderBoardから大きくShakeupしたため、革新的な解法ではないと思います。</p>\n<p>主に実施した内容としては以下になります。</p>\n<ol>\n<li>TimeSeriesのデータはAggregateとDeepAutoEncoderを利用</li>\n<li>特徴量エンジニアリング</li>\n<li>目的変数のない行は削除</li>\n<li>評価指標の3つの境界値もOptunaで最適化</li>\n<li>SeedAveragingの実施。(ただしシングルモデルの方が性能が良かった。)</li>\n<li>LightGBMを採用したが、XGBoostとかCatBoostとかTabNetも試した。</li>\n</ol>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder</a></p>\n<p>追記<br>\n添付したファイルはGoogleColabで実行していたファイルです。Seed値は手動で切り替えて最終提出でSeedAverageをしています。<br>\n最終提出ではGoogleColabで得たモデルと閾値をベースに、Seedで多数決をとって提出しています。<a href=\"https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission</a></p>",
  "messages": [
    {
      "id": 3077753,
      "postDate": "2024-12-21T11:08:13.303Z",
      "content": "<p>---@English<br>\nHere is the solution I would like to share.</p>\n<p>Since there was a significant shakeup on the Public Leaderboard, I don't believe this is a groundbreaking solution.</p>\n<p>The main approaches implemented are as follows:</p>\n<p>Aggregating and using DeepAutoEncoder for TimeSeries data<br>\nFeature Engineering<br>\nRemoving rows without the target variable<br>\nOptimizing the three boundary values for the evaluation metric using Optuna<br>\nApplying Seed Averaging (although the single model performed better)<br>\nUsing LightGBM, while also experimenting with XGBoost, CatBoost, and TabNet.</p>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder</a></p>\n<p>Addition:<br>\nThe attached file is a script that was executed on Google Colab. The seed values were manually switched, and a Seed Average was applied for the final submission.<br>\nFor the final submission, we based it on the model and thresholds obtained from Google Colab, and used a majority vote across different seeds for submission.</p>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission</a></p>\n<p>----@JP<br>\nソリューションを共有します。</p>\n<p>Public LeaderBoardから大きくShakeupしたため、革新的な解法ではないと思います。</p>\n<p>主に実施した内容としては以下になります。</p>\n<ol>\n<li>TimeSeriesのデータはAggregateとDeepAutoEncoderを利用</li>\n<li>特徴量エンジニアリング</li>\n<li>目的変数のない行は削除</li>\n<li>評価指標の3つの境界値もOptunaで最適化</li>\n<li>SeedAveragingの実施。(ただしシングルモデルの方が性能が良かった。)</li>\n<li>LightGBMを採用したが、XGBoostとかCatBoostとかTabNetも試した。</li>\n</ol>\n<p><a href=\"https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder</a></p>\n<p>追記<br>\n添付したファイルはGoogleColabで実行していたファイルです。Seed値は手動で切り替えて最終提出でSeedAverageをしています。<br>\n最終提出ではGoogleColabで得たモデルと閾値をベースに、Seedで多数決をとって提出しています。<a href=\"https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\" target=\"_blank\">https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission</a></p>",
      "rawMarkdown": "---@English\nHere is the solution I would like to share.\n\nSince there was a significant shakeup on the Public Leaderboard, I don't believe this is a groundbreaking solution.\n\nThe main approaches implemented are as follows:\n\nAggregating and using DeepAutoEncoder for TimeSeries data\nFeature Engineering\nRemoving rows without the target variable\nOptimizing the three boundary values for the evaluation metric using Optuna\nApplying Seed Averaging (although the single model performed better)\nUsing LightGBM, while also experimenting with XGBoost, CatBoost, and TabNet.\n\nhttps://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\n\nAddition:\nThe attached file is a script that was executed on Google Colab. The seed values were manually switched, and a Seed Average was applied for the final submission.\nFor the final submission, we based it on the model and thresholds obtained from Google Colab, and used a majority vote across different seeds for submission.\n\nhttps://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\n\n----@JP\nソリューションを共有します。\n\nPublic LeaderBoardから大きくShakeupしたため、革新的な解法ではないと思います。\n\n主に実施した内容としては以下になります。\n1. TimeSeriesのデータはAggregateとDeepAutoEncoderを利用\n2. 特徴量エンジニアリング\n3. 目的変数のない行は削除\n4. 評価指標の3つの境界値もOptunaで最適化\n5. SeedAveragingの実施。(ただしシングルモデルの方が性能が良かった。)\n6. LightGBMを採用したが、XGBoostとかCatBoostとかTabNetも試した。\n\nhttps://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\n\n追記\n添付したファイルはGoogleColabで実行していたファイルです。Seed値は手動で切り替えて最終提出でSeedAverageをしています。\n最終提出ではGoogleColabで得たモデルと閾値をベースに、Seedで多数決をとって提出しています。https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\n\n",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3077753": "---@English\nHere is the solution I would like to share.\n\nSince there was a significant shakeup on the Public Leaderboard, I don't believe this is a groundbreaking solution.\n\nThe main approaches implemented are as follows:\n\nAggregating and using DeepAutoEncoder for TimeSeries data\nFeature Engineering\nRemoving rows without the target variable\nOptimizing the three boundary values for the evaluation metric using Optuna\nApplying Seed Averaging (although the single model performed better)\nUsing LightGBM, while also experimenting with XGBoost, CatBoost, and TabNet.\n\nhttps://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\n\nAddition:\nThe attached file is a script that was executed on Google Colab. The seed values were manually switched, and a Seed Average was applied for the final submission.\nFor the final submission, we based it on the model and thresholds obtained from Google Colab, and used a majority vote across different seeds for submission.\n\nhttps://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\n\n----@JP\nソリューションを共有します。\n\nPublic LeaderBoardから大きくShakeupしたため、革新的な解法ではないと思います。\n\n主に実施した内容としては以下になります。\n1. TimeSeriesのデータはAggregateとDeepAutoEncoderを利用\n2. 特徴量エンジニアリング\n3. 目的変数のない行は削除\n4. 評価指標の3つの境界値もOptunaで最適化\n5. SeedAveragingの実施。(ただしシングルモデルの方が性能が良かった。)\n6. LightGBMを採用したが、XGBoostとかCatBoostとかTabNetも試した。\n\nhttps://www.kaggle.com/code/pegasus27/solution-lightgbm-deepautoencoder\n\n追記\n添付したファイルはGoogleColabで実行していたファイルです。Seed値は手動で切り替えて最終提出でSeedAverageをしています。\n最終提出ではGoogleColabで得たモデルと閾値をベースに、Seedで多数決をとって提出しています。https://www.kaggle.com/code/pegasus27/cmi-pi-lightgbm-deep-auto-encoder-submission\n\n"
  }
}