{
  "id": 508167,
  "title": "Top 69 solution.",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/508167",
  "author_name": "",
  "post_date": "2024-05-28T13:52:26.342081400Z",
  "votes": 6,
  "comment_count": 1,
  "views": 0,
  "content": "<p><a href=\"https://www.kaggle.com/code/skrrydg/homecreditinput\" target=\"_blank\">https://www.kaggle.com/code/skrrydg/homecreditinput</a><br>\n<a href=\"https://www.kaggle.com/code/skrrydg/homecreditmodelpredict\" target=\"_blank\">https://www.kaggle.com/code/skrrydg/homecreditmodelpredict</a><br>\n<a href=\"https://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs\" target=\"_blank\">https://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs</a></p>\n<p>What interesting you can find in github repo?</p>\n<ol>\n<li>The framework of data processing, which able to generate more than 3k features on whole dataset without oom.(see 'input' and 'preprocessor' directories)</li>\n<li>3 ways to restore date decision('date_decision_restorer' directory) </li>\n<li>Tens different steps of data prtocessing ('preprocessor/steps' directory)</li>\n<li>Memory effective way to learn lightgbm. 1k+ features without oom ('lightgbm' directory)</li>\n<li>Several approaches for feature selection, like psi feature selection ('feature_selection' directory)</li>\n</ol>\n<p>I have a lot of numbers, but want to share my best single model result: <br>\nPublic: 0.598<br>\nPrivate: 0.522</p>\n<p>P.S. I forgot to make private my github repo, so it was easy to find it  :D</p>",
  "messages": [
    {
      "id": "2841277",
      "postDate": "05/28/2024 13:52:26",
      "content": "<p><a href=\"https://www.kaggle.com/code/skrrydg/homecreditinput\" target=\"_blank\">https://www.kaggle.com/code/skrrydg/homecreditinput</a><br>\n<a href=\"https://www.kaggle.com/code/skrrydg/homecreditmodelpredict\" target=\"_blank\">https://www.kaggle.com/code/skrrydg/homecreditmodelpredict</a><br>\n<a href=\"https://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs\" target=\"_blank\">https://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs</a></p>\n<p>What interesting you can find in github repo?</p>\n<ol>\n<li>The framework of data processing, which able to generate more than 3k features on whole dataset without oom.(see 'input' and 'preprocessor' directories)</li>\n<li>3 ways to restore date decision('date_decision_restorer' directory) </li>\n<li>Tens different steps of data prtocessing ('preprocessor/steps' directory)</li>\n<li>Memory effective way to learn lightgbm. 1k+ features without oom ('lightgbm' directory)</li>\n<li>Several approaches for feature selection, like psi feature selection ('feature_selection' directory)</li>\n</ol>\n<p>I have a lot of numbers, but want to share my best single model result: <br>\nPublic: 0.598<br>\nPrivate: 0.522</p>\n<p>P.S. I forgot to make private my github repo, so it was easy to find it  :D</p>",
      "rawMarkdown": "https://www.kaggle.com/code/skrrydg/homecreditinput\nhttps://www.kaggle.com/code/skrrydg/homecreditmodelpredict\nhttps://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs\n\n\nWhat interesting you can find in github repo?\n1. The framework of data processing, which able to generate more than 3k features on whole dataset without oom.(see 'input' and 'preprocessor' directories)\n2. 3 ways to restore date decision('date_decision_restorer' directory) \n3. Tens different steps of data prtocessing ('preprocessor/steps' directory)\n4. Memory effective way to learn lightgbm. 1k+ features without oom ('lightgbm' directory)\n5. Several approaches for feature selection, like psi feature selection ('feature_selection' directory)\n\nI have a lot of numbers, but want to share my best single model result: \nPublic: 0.598\nPrivate: 0.522\n\nP.S. I forgot to make private my github repo, so it was easy to find it  :D",
      "votes": null
    },
    {
      "id": "2842182",
      "postDate": "05/28/2024 23:48:52",
      "content": "<p>вмк дед забрал свое </p>",
      "rawMarkdown": "вмк дед забрал свое",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2842182,
      "author_name": "oxygen22831",
      "author_url": "",
      "post_date": "05/28/2024 23:48:52",
      "content": "<p>вмк дед забрал свое </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2841277": "https://www.kaggle.com/code/skrrydg/homecreditinput\nhttps://www.kaggle.com/code/skrrydg/homecreditmodelpredict\nhttps://github.com/skrydg/kaggle_home_credit_risk_model_stability/tree/master/src/kaggle_home_credit_risk_model_stability/libs\n\n\nWhat interesting you can find in github repo?\n1. The framework of data processing, which able to generate more than 3k features on whole dataset without oom.(see 'input' and 'preprocessor' directories)\n2. 3 ways to restore date decision('date_decision_restorer' directory) \n3. Tens different steps of data prtocessing ('preprocessor/steps' directory)\n4. Memory effective way to learn lightgbm. 1k+ features without oom ('lightgbm' directory)\n5. Several approaches for feature selection, like psi feature selection ('feature_selection' directory)\n\nI have a lot of numbers, but want to share my best single model result: \nPublic: 0.598\nPrivate: 0.522\n\nP.S. I forgot to make private my github repo, so it was easy to find it  :D",
    "2842182": "вмк дед забрал свое"
  },
  "source": "meta"
}