{
  "id": 59735,
  "title": "dataset imbalance learn ",
  "url": "/competitions/avazu-ctr-prediction/discussion/59735",
  "author_name": "",
  "post_date": "2018-06-26T15:45:16.508606200Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>I would like to balance the dataset meaning consider 50% clicks and 50% non-clicks for training the model.\nI have read about the resampling techniques and also about the python imblearn library.\nAny suggestions or approach to overcome the stated problem will be appreciated!</p>\n\n<p>Thanks.</p>",
  "messages": [
    {
      "id": "348374",
      "postDate": "06/26/2018 15:45:16",
      "content": "<p>Hello,</p>\n\n<p>I would like to balance the dataset meaning consider 50% clicks and 50% non-clicks for training the model.\nI have read about the resampling techniques and also about the python imblearn library.\nAny suggestions or approach to overcome the stated problem will be appreciated!</p>\n\n<p>Thanks.</p>",
      "rawMarkdown": "Hello,\n\nI would like to balance the dataset meaning consider 50% clicks and 50% non-clicks for training the model.\nI have read about the resampling techniques and also about the python imblearn library.\nAny suggestions or approach to overcome the stated problem will be appreciated!\n\nThanks.",
      "votes": null
    },
    {
      "id": "354235",
      "postDate": "07/09/2018 06:03:49",
      "content": "<p>Undersampling often leads to overfitting. You might also consider oversampling techniques; check out this kernel: <a href=\"https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail\">https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail</a></p>\n\n<p>Personally, I've found Gradient Boosted Machines (GBMs) to be quite resilient to class imbalance, and that combining them with oversampling / undersampling techniques <em>always</em> yields worse results than when using GBMs alone.</p>",
      "rawMarkdown": "Undersampling often leads to overfitting. You might also consider oversampling techniques; check out this kernel: https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail\n\nPersonally, I've found Gradient Boosted Machines (GBMs) to be quite resilient to class imbalance, and that combining them with oversampling / undersampling techniques *always* yields worse results than when using GBMs alone.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 354235,
      "author_name": "ponadto",
      "author_url": "",
      "post_date": "07/09/2018 06:03:49",
      "content": "<p>Undersampling often leads to overfitting. You might also consider oversampling techniques; check out this kernel: <a href=\"https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail\">https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail</a></p>\n\n<p>Personally, I've found Gradient Boosted Machines (GBMs) to be quite resilient to class imbalance, and that combining them with oversampling / undersampling techniques <em>always</em> yields worse results than when using GBMs alone.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "348374": "Hello,\n\nI would like to balance the dataset meaning consider 50% clicks and 50% non-clicks for training the model.\nI have read about the resampling techniques and also about the python imblearn library.\nAny suggestions or approach to overcome the stated problem will be appreciated!\n\nThanks.",
    "354235": "Undersampling often leads to overfitting. You might also consider oversampling techniques; check out this kernel: https://www.kaggle.com/gargmanish/how-to-handle-imbalance-data-study-in-detail\n\nPersonally, I've found Gradient Boosted Machines (GBMs) to be quite resilient to class imbalance, and that combining them with oversampling / undersampling techniques *always* yields worse results than when using GBMs alone."
  },
  "source": "meta"
}