{
  "id": 56116,
  "title": "What about a technical topic à la fin",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/56116",
  "author_name": "Toaru",
  "post_date": "2018-05-06T11:12:18.607000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Did anyone tried entity embedding in this competition?\nSince the feature importance of categorical variables are all very high, I wondered if a entity embedding can improve the score or not.</p>\n\n<p>I have tried to get a vector representation of every cate variable in the light of this paper <a href=\"https://arxiv.org/pdf/1604.06737.pdf\">https://arxiv.org/pdf/1604.06737.pdf</a> and used the NN struct of <a href=\"https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s\">https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s</a>\nbut the result seems not very good.</p>\n\n<p>Besides NN, is there any other method to generate the vector representation of cate variable?</p>",
  "messages": [
    {
      "id": 323834,
      "postDate": "2018-05-06T11:12:18.607Z",
      "content": "<p>Did anyone tried entity embedding in this competition?\nSince the feature importance of categorical variables are all very high, I wondered if a entity embedding can improve the score or not.</p>\n\n<p>I have tried to get a vector representation of every cate variable in the light of this paper <a href=\"https://arxiv.org/pdf/1604.06737.pdf\">https://arxiv.org/pdf/1604.06737.pdf</a> and used the NN struct of <a href=\"https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s\">https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s</a>\nbut the result seems not very good.</p>\n\n<p>Besides NN, is there any other method to generate the vector representation of cate variable?</p>",
      "rawMarkdown": "Did anyone tried entity embedding in this competition?\nSince the feature importance of categorical variables are all very high, I wondered if a entity embedding can improve the score or not.\n\nI have tried to get a vector representation of every cate variable in the light of this paper https://arxiv.org/pdf/1604.06737.pdf and used the NN struct of https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s\nbut the result seems not very good.\n\nBesides NN, is there any other method to generate the vector representation of cate variable?\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "323834": "Did anyone tried entity embedding in this competition?\nSince the feature importance of categorical variables are all very high, I wondered if a entity embedding can improve the score or not.\n\nI have tried to get a vector representation of every cate variable in the light of this paper https://arxiv.org/pdf/1604.06737.pdf and used the NN struct of https://www.kaggle.com/whitebird/mercari-price-3rd-0-3905-cv-at-pb-in-3300-s\nbut the result seems not very good.\n\nBesides NN, is there any other method to generate the vector representation of cate variable?\n"
  }
}