{
  "id": 58395,
  "title": "Mean Target Value (MTV) based feature engineering have performed better than one hot encoding (OHE). Any reason why?",
  "url": "/competitions/avito-demand-prediction/discussion/58395",
  "author_name": "",
  "post_date": "2018-06-07T09:37:48.006574500Z",
  "votes": 12,
  "comment_count": 2,
  "views": 0,
  "content": "<p>We have lot of ways of converting categorical variables to numerical once. Few of the techniques are:</p>\n\n<ol>\n<li>Label Encoding</li>\n<li>One Hot Encoding</li>\n<li>Mean Target Value (based on deal_probability in this competition)</li>\n</ol>\n\n<p>I want to understand the effectiveness of using technique #3 compared to #1 and #2. I find that MTV performed better than others.</p>\n\n<p>Thanks in advance</p>",
  "messages": [
    {
      "id": "339654",
      "postDate": "06/07/2018 09:37:48",
      "content": "<p>We have lot of ways of converting categorical variables to numerical once. Few of the techniques are:</p>\n\n<ol>\n<li>Label Encoding</li>\n<li>One Hot Encoding</li>\n<li>Mean Target Value (based on deal_probability in this competition)</li>\n</ol>\n\n<p>I want to understand the effectiveness of using technique #3 compared to #1 and #2. I find that MTV performed better than others.</p>\n\n<p>Thanks in advance</p>",
      "rawMarkdown": "We have lot of ways of converting categorical variables to numerical once. Few of the techniques are:\n\n1. Label Encoding\n2. One Hot Encoding\n3. Mean Target Value (based on deal_probability in this competition)\n\nI want to understand the effectiveness of using technique #3 compared to #1 and #2. I find that MTV performed better than others.\n\nThanks in advance",
      "votes": null
    },
    {
      "id": "339664",
      "postDate": "06/07/2018 09:57:33",
      "content": "<p>Using technique #1 you need a tree-based and boosting learner,  Using technique #2 is more general way, and technique #3 is implemented by catboost, and it outperforms the others. Correct me if I'm wrong.</p>",
      "rawMarkdown": "Using technique #1 you need a tree-based and boosting learner,  Using technique #2 is more general way, and technique #3 is implemented by catboost, and it outperforms the others. Correct me if I'm wrong.",
      "votes": null
    },
    {
      "id": "339768",
      "postDate": "06/07/2018 15:45:01",
      "content": "<p>Cheers for the explanation. I read more about it and found that technique #3 is termed as Mean Target Value (MTV) based feature engineering. As you mentioned this technique is implemented in catboost but no such option is provided in sklearn.</p>\n\n<p>-1 for my topic is demotivating. But I will continue to ask stupid questions. :)</p>",
      "rawMarkdown": "Cheers for the explanation. I read more about it and found that technique #3 is termed as Mean Target Value (MTV) based feature engineering. As you mentioned this technique is implemented in catboost but no such option is provided in sklearn.\n\n-1 for my topic is demotivating. But I will continue to ask stupid questions. :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 339664,
      "author_name": "marcuslin",
      "author_url": "",
      "post_date": "06/07/2018 09:57:33",
      "content": "<p>Using technique #1 you need a tree-based and boosting learner,  Using technique #2 is more general way, and technique #3 is implemented by catboost, and it outperforms the others. Correct me if I'm wrong.</p>",
      "votes": null,
      "replies": [
        {
          "id": 339768,
          "author_name": "abdul0807",
          "author_url": "",
          "post_date": "06/07/2018 15:45:01",
          "content": "<p>Cheers for the explanation. I read more about it and found that technique #3 is termed as Mean Target Value (MTV) based feature engineering. As you mentioned this technique is implemented in catboost but no such option is provided in sklearn.</p>\n\n<p>-1 for my topic is demotivating. But I will continue to ask stupid questions. :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "339654": "We have lot of ways of converting categorical variables to numerical once. Few of the techniques are:\n\n1. Label Encoding\n2. One Hot Encoding\n3. Mean Target Value (based on deal_probability in this competition)\n\nI want to understand the effectiveness of using technique #3 compared to #1 and #2. I find that MTV performed better than others.\n\nThanks in advance",
    "339664": "Using technique #1 you need a tree-based and boosting learner,  Using technique #2 is more general way, and technique #3 is implemented by catboost, and it outperforms the others. Correct me if I'm wrong.",
    "339768": "Cheers for the explanation. I read more about it and found that technique #3 is termed as Mean Target Value (MTV) based feature engineering. As you mentioned this technique is implemented in catboost but no such option is provided in sklearn.\n\n-1 for my topic is demotivating. But I will continue to ask stupid questions. :)"
  },
  "source": "meta"
}