{
  "id": 24810,
  "title": "Over Sampling",
  "url": "/competitions/bosch-production-line-performance/discussion/24810",
  "author_name": "",
  "post_date": "2016-10-27T06:48:02.650Z",
  "votes": null,
  "comment_count": 2,
  "views": 614,
  "content": "<p>Hi,\nAs I am trying to perform Xgboost first time, I am just wondering how much response percent I should include in my training dataset. I would admire if anyone can help me in understanding how oversampling should be done here as there is 6789 1's out of 1183747 rows. (approx 1% ) and I cannot train my model with just 1% response rate.</p>",
  "messages": [
    {
      "id": "141504",
      "postDate": "10/27/2016 06:48:02",
      "content": "<p>Hi,\nAs I am trying to perform Xgboost first time, I am just wondering how much response percent I should include in my training dataset. I would admire if anyone can help me in understanding how oversampling should be done here as there is 6789 1's out of 1183747 rows. (approx 1% ) and I cannot train my model with just 1% response rate.</p>",
      "rawMarkdown": "Hi,\r\nAs I am trying to perform Xgboost first time, I am just wondering how much response percent I should include in my training dataset. I would admire if anyone can help me in understanding how oversampling should be done here as there is 6789 1's out of 1183747 rows. (approx 1% ) and I cannot train my model with just 1% response rate.",
      "votes": null
    },
    {
      "id": "141526",
      "postDate": "10/27/2016 09:05:22",
      "content": "<p>You can train xgboost with 0.5% positives, and 99.5% negatives, it is just that the objective function cannot be accuracy. Trying using AUC or make a customized objective function of MCC. There are kernels which have done that already. </p>",
      "rawMarkdown": "You can train xgboost with 0.5% positives, and 99.5% negatives, it is just that the objective function cannot be accuracy. Trying using AUC or make a customized objective function of MCC. There are kernels which have done that already.",
      "votes": null
    },
    {
      "id": "141733",
      "postDate": "10/28/2016 04:43:18",
      "content": "<p>Okay, thanks a lot for this help.\nReally appreciated :) </p>",
      "rawMarkdown": "Okay, thanks a lot for this help.\r\nReally appreciated :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 141526,
      "author_name": "leexiongan",
      "author_url": "",
      "post_date": "10/27/2016 09:05:22",
      "content": "<p>You can train xgboost with 0.5% positives, and 99.5% negatives, it is just that the objective function cannot be accuracy. Trying using AUC or make a customized objective function of MCC. There are kernels which have done that already. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 141733,
      "author_name": "anmoljain",
      "author_url": "",
      "post_date": "10/28/2016 04:43:18",
      "content": "<p>Okay, thanks a lot for this help.\nReally appreciated :) </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "141504": "Hi,\r\nAs I am trying to perform Xgboost first time, I am just wondering how much response percent I should include in my training dataset. I would admire if anyone can help me in understanding how oversampling should be done here as there is 6789 1's out of 1183747 rows. (approx 1% ) and I cannot train my model with just 1% response rate.",
    "141526": "You can train xgboost with 0.5% positives, and 99.5% negatives, it is just that the objective function cannot be accuracy. Trying using AUC or make a customized objective function of MCC. There are kernels which have done that already.",
    "141733": "Okay, thanks a lot for this help.\r\nReally appreciated :)"
  },
  "source": "meta"
}