{
  "id": 66675,
  "title": "Could we use Classification ?",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/66675",
  "author_name": "",
  "post_date": "2018-09-24T06:43:09.661018700Z",
  "votes": 11,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Just realeased 2 new kernels:</p>\n\n<ul>\n<li>The first one is a <a href=\"https://www.kaggle.com/ogrellier/create-extracted-json-fields-dataset\">utility kernel</a> that yields readily exctracted json fields for train and test data for those of you who don't want to spend time doing it. </li>\n<li>The second one is more about <a href=\"https://www.kaggle.com/ogrellier/using-classification-for-predictions\">classification and provides a 1.6667 LB model</a></li>\n</ul>\n\n<p>Our biggest issue I think with this data is to get the zero revenue transactions right during regression. So I was wondering if classifying non-zero transactions first would help. The answer is <a href=\"https://www.kaggle.com/ogrellier/using-classification-for-predictions\">here</a>. Nothing fancy just the raw features with simple models.</p>\n\n<p>Hope you'll find this useful ;-)</p>",
  "messages": [
    {
      "id": "392654",
      "postDate": "09/24/2018 06:43:09",
      "content": "<p>Just realeased 2 new kernels:</p>\n\n<ul>\n<li>The first one is a <a href=\"https://www.kaggle.com/ogrellier/create-extracted-json-fields-dataset\">utility kernel</a> that yields readily exctracted json fields for train and test data for those of you who don't want to spend time doing it. </li>\n<li>The second one is more about <a href=\"https://www.kaggle.com/ogrellier/using-classification-for-predictions\">classification and provides a 1.6667 LB model</a></li>\n</ul>\n\n<p>Our biggest issue I think with this data is to get the zero revenue transactions right during regression. So I was wondering if classifying non-zero transactions first would help. The answer is <a href=\"https://www.kaggle.com/ogrellier/using-classification-for-predictions\">here</a>. Nothing fancy just the raw features with simple models.</p>\n\n<p>Hope you'll find this useful ;-)</p>",
      "rawMarkdown": "Just realeased 2 new kernels:\n\n - The first one is a [utility kernel](https://www.kaggle.com/ogrellier/create-extracted-json-fields-dataset) that yields readily exctracted json fields for train and test data for those of you who don't want to spend time doing it. \n - The second one is more about [classification and provides a 1.6667 LB model](https://www.kaggle.com/ogrellier/using-classification-for-predictions)\n\nOur biggest issue I think with this data is to get the zero revenue transactions right during regression. So I was wondering if classifying non-zero transactions first would help. The answer is [here](https://www.kaggle.com/ogrellier/using-classification-for-predictions). Nothing fancy just the raw features with simple models.\n\nHope you'll find this useful ;-)",
      "votes": null
    },
    {
      "id": "392662",
      "postDate": "09/24/2018 06:50:01",
      "content": "<p>My intuition is also in this direction. First classify 0 or more.</p>",
      "rawMarkdown": "My intuition is also in this direction. First classify 0 or more.",
      "votes": null
    },
    {
      "id": "392828",
      "postDate": "09/24/2018 12:47:52",
      "content": "<p>I had this idea too: first implement a classification model that has high specificity (i.e., a model in which we can be certain about true negatives / 'no money' class). Then, apply a regression model to the rest. On my to-do list!</p>",
      "rawMarkdown": "I had this idea too: first implement a classification model that has high specificity (i.e., a model in which we can be certain about true negatives / 'no money' class). Then, apply a regression model to the rest. On my to-do list!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 392662,
      "author_name": "abosol",
      "author_url": "",
      "post_date": "09/24/2018 06:50:01",
      "content": "<p>My intuition is also in this direction. First classify 0 or more.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 392828,
      "author_name": "nnnnick",
      "author_url": "",
      "post_date": "09/24/2018 12:47:52",
      "content": "<p>I had this idea too: first implement a classification model that has high specificity (i.e., a model in which we can be certain about true negatives / 'no money' class). Then, apply a regression model to the rest. On my to-do list!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "392654": "Just realeased 2 new kernels:\n\n - The first one is a [utility kernel](https://www.kaggle.com/ogrellier/create-extracted-json-fields-dataset) that yields readily exctracted json fields for train and test data for those of you who don't want to spend time doing it. \n - The second one is more about [classification and provides a 1.6667 LB model](https://www.kaggle.com/ogrellier/using-classification-for-predictions)\n\nOur biggest issue I think with this data is to get the zero revenue transactions right during regression. So I was wondering if classifying non-zero transactions first would help. The answer is [here](https://www.kaggle.com/ogrellier/using-classification-for-predictions). Nothing fancy just the raw features with simple models.\n\nHope you'll find this useful ;-)",
    "392662": "My intuition is also in this direction. First classify 0 or more.",
    "392828": "I had this idea too: first implement a classification model that has high specificity (i.e., a model in which we can be certain about true negatives / 'no money' class). Then, apply a regression model to the rest. On my to-do list!"
  },
  "source": "meta"
}