{
  "id": 129774,
  "title": "Structured approach to Pipeline and Grid Search",
  "url": "/competitions/cat-in-the-dat-ii/discussion/129774",
  "author_name": "",
  "post_date": "2020-02-10T17:20:42.713533200Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I would like to share a kernel I made to illustrate a structured approach to building a pipeline for feature preprocessing and model evaluation. <a href=\"https://www.kaggle.com/esasiela/cat-in-the-dat-ii-grid-search\">Cat in the Dat II - Grid Search and Pipeline</a></p>\n\n<p>Beyond simply creating a pipeline, this kernel also tweaks the Scikit-learn get_feature_names() capability, which is only partially implemented in the default sklearn version on Kaggle.  This allows for clean traceability of the final transformed features backwards through the pipeline.</p>\n\n<p>The implementation only includes the ordinal features as a reference, so it does not score well (AUC: 0.71289 on a validation hold out set).  The purpose of sharing the kernel is more for the structured approach that can be applied to any solution, rather than showing how to score high on this one.</p>\n\n<p>This is my first public kernel, so I would love to hear any thoughts you have.</p>",
  "messages": [
    {
      "id": "741467",
      "postDate": "02/10/2020 17:20:42",
      "content": "<p>I would like to share a kernel I made to illustrate a structured approach to building a pipeline for feature preprocessing and model evaluation. <a href=\"https://www.kaggle.com/esasiela/cat-in-the-dat-ii-grid-search\">Cat in the Dat II - Grid Search and Pipeline</a></p>\n\n<p>Beyond simply creating a pipeline, this kernel also tweaks the Scikit-learn get_feature_names() capability, which is only partially implemented in the default sklearn version on Kaggle.  This allows for clean traceability of the final transformed features backwards through the pipeline.</p>\n\n<p>The implementation only includes the ordinal features as a reference, so it does not score well (AUC: 0.71289 on a validation hold out set).  The purpose of sharing the kernel is more for the structured approach that can be applied to any solution, rather than showing how to score high on this one.</p>\n\n<p>This is my first public kernel, so I would love to hear any thoughts you have.</p>",
      "rawMarkdown": "I would like to share a kernel I made to illustrate a structured approach to building a pipeline for feature preprocessing and model evaluation. [Cat in the Dat II - Grid Search and Pipeline](https://www.kaggle.com/esasiela/cat-in-the-dat-ii-grid-search)\n\nBeyond simply creating a pipeline, this kernel also tweaks the Scikit-learn get_feature_names() capability, which is only partially implemented in the default sklearn version on Kaggle.  This allows for clean traceability of the final transformed features backwards through the pipeline.\n\nThe implementation only includes the ordinal features as a reference, so it does not score well (AUC: 0.71289 on a validation hold out set).  The purpose of sharing the kernel is more for the structured approach that can be applied to any solution, rather than showing how to score high on this one.\n\nThis is my first public kernel, so I would love to hear any thoughts you have.",
      "votes": null
    },
    {
      "id": "741496",
      "postDate": "02/10/2020 18:07:58",
      "content": "<p>Very nice!! Thanks for sharing</p>",
      "rawMarkdown": "Very nice!! Thanks for sharing",
      "votes": null
    },
    {
      "id": "741657",
      "postDate": "02/10/2020 21:29:50",
      "content": "<p>I've already seen your excellent (grid search) Kaggle Notebook. </p>",
      "rawMarkdown": "I've already seen your excellent (grid search) Kaggle Notebook.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 741496,
      "author_name": "victormenuzzo",
      "author_url": "",
      "post_date": "02/10/2020 18:07:58",
      "content": "<p>Very nice!! Thanks for sharing</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 741657,
      "author_name": "mpwolke",
      "author_url": "",
      "post_date": "02/10/2020 21:29:50",
      "content": "<p>I've already seen your excellent (grid search) Kaggle Notebook. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "741467": "I would like to share a kernel I made to illustrate a structured approach to building a pipeline for feature preprocessing and model evaluation. [Cat in the Dat II - Grid Search and Pipeline](https://www.kaggle.com/esasiela/cat-in-the-dat-ii-grid-search)\n\nBeyond simply creating a pipeline, this kernel also tweaks the Scikit-learn get_feature_names() capability, which is only partially implemented in the default sklearn version on Kaggle.  This allows for clean traceability of the final transformed features backwards through the pipeline.\n\nThe implementation only includes the ordinal features as a reference, so it does not score well (AUC: 0.71289 on a validation hold out set).  The purpose of sharing the kernel is more for the structured approach that can be applied to any solution, rather than showing how to score high on this one.\n\nThis is my first public kernel, so I would love to hear any thoughts you have.",
    "741496": "Very nice!! Thanks for sharing",
    "741657": "I've already seen your excellent (grid search) Kaggle Notebook."
  },
  "source": "meta"
}