{
  "id": 67484,
  "title": "Improving just LB",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/67484",
  "author_name": "",
  "post_date": "2018-10-03T07:34:29.102646500Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Trust your CV scores in this competition. Multiply your predictions by 0.97 and your LB will improve(which will be disastrous in the private LB I think). Maybe this suggests that public data consists of even a higher percentage of zeros than our train data and  my model is performing badly on the revenues which are greater than zero.</p>",
  "messages": [
    {
      "id": "397830",
      "postDate": "10/03/2018 07:34:29",
      "content": "<p>Trust your CV scores in this competition. Multiply your predictions by 0.97 and your LB will improve(which will be disastrous in the private LB I think). Maybe this suggests that public data consists of even a higher percentage of zeros than our train data and  my model is performing badly on the revenues which are greater than zero.</p>",
      "rawMarkdown": "Trust your CV scores in this competition. Multiply your predictions by 0.97 and your LB will improve(which will be disastrous in the private LB I think). Maybe this suggests that public data consists of even a higher percentage of zeros than our train data and  my model is performing badly on the revenues which are greater than zero.",
      "votes": null
    },
    {
      "id": "398354",
      "postDate": "10/04/2018 01:41:43",
      "content": "<p>why multiply 0.97? adjust to zero percentage(?)</p>",
      "rawMarkdown": "why multiply 0.97? adjust to zero percentage(?)",
      "votes": null
    },
    {
      "id": "398415",
      "postDate": "10/04/2018 04:57:58",
      "content": "<p>I mean any number around to reduce the non-zero revenues from getting too big(This maybe the case with the public test part, in which there is hardly any transaction with high log revenue). Maybe my model is giving higher log revenues to these non-zero ones, so its helping to adjust them. </p>",
      "rawMarkdown": "I mean any number around to reduce the non-zero revenues from getting too big(This maybe the case with the public test part, in which there is hardly any transaction with high log revenue). Maybe my model is giving higher log revenues to these non-zero ones, so its helping to adjust them.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 398354,
      "author_name": "chocozzz",
      "author_url": "",
      "post_date": "10/04/2018 01:41:43",
      "content": "<p>why multiply 0.97? adjust to zero percentage(?)</p>",
      "votes": null,
      "replies": [
        {
          "id": 398415,
          "author_name": "nikhilmishradev",
          "author_url": "",
          "post_date": "10/04/2018 04:57:58",
          "content": "<p>I mean any number around to reduce the non-zero revenues from getting too big(This maybe the case with the public test part, in which there is hardly any transaction with high log revenue). Maybe my model is giving higher log revenues to these non-zero ones, so its helping to adjust them. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "397830": "Trust your CV scores in this competition. Multiply your predictions by 0.97 and your LB will improve(which will be disastrous in the private LB I think). Maybe this suggests that public data consists of even a higher percentage of zeros than our train data and  my model is performing badly on the revenues which are greater than zero.",
    "398354": "why multiply 0.97? adjust to zero percentage(?)",
    "398415": "I mean any number around to reduce the non-zero revenues from getting too big(This maybe the case with the public test part, in which there is hardly any transaction with high log revenue). Maybe my model is giving higher log revenues to these non-zero ones, so its helping to adjust them."
  },
  "source": "meta"
}