{
  "id": 100711,
  "title": "Mixing the Old and New Dataset",
  "url": "/competitions/aptos2019-blindness-detection/discussion/100711",
  "author_name": "",
  "post_date": "2019-07-20T10:53:03.130173Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello there, I want to know if it is normal to mix the old and new dataset in a balanced way? are the distribution of the old and new data the same? Will the Model Generalize well if I do this Mixup? Thank you very much.</p>",
  "messages": [
    {
      "id": "580550",
      "postDate": "07/20/2019 10:53:03",
      "content": "<p>Hello there, I want to know if it is normal to mix the old and new dataset in a balanced way? are the distribution of the old and new data the same? Will the Model Generalize well if I do this Mixup? Thank you very much.</p>",
      "rawMarkdown": "Hello there, I want to know if it is normal to mix the old and new dataset in a balanced way? are the distribution of the old and new data the same? Will the Model Generalize well if I do this Mixup? Thank you very much.",
      "votes": null
    },
    {
      "id": "582037",
      "postDate": "07/22/2019 17:29:06",
      "content": "<p><a href=\"/mxcsyounes\">@mxcsyounes</a> Rather than mixing the 2 datasets, which will make the task of making the cv troublesome and maybe less dependable. The following method shared by <a href=\"/drhabib\">@drhabib</a> has seemed to work, (<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005</a>) : Pretraining the model on 2015 dataset while using 2019 as validation, then finetuning the model using 2019 dataset for training and a 2019 subset as CV.</p>",
      "rawMarkdown": "mxcsyounes Rather than mixing the 2 datasets, which will make the task of making the cv troublesome and maybe less dependable. The following method shared by @drhabib has seemed to work, (https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005) : Pretraining the model on 2015 dataset while using 2019 as validation, then finetuning the model using 2019 dataset for training and a 2019 subset as CV.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 582037,
      "author_name": "pranavmahajan725",
      "author_url": "",
      "post_date": "07/22/2019 17:29:06",
      "content": "<p><a href=\"/mxcsyounes\">@mxcsyounes</a> Rather than mixing the 2 datasets, which will make the task of making the cv troublesome and maybe less dependable. The following method shared by <a href=\"/drhabib\">@drhabib</a> has seemed to work, (<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005</a>) : Pretraining the model on 2015 dataset while using 2019 as validation, then finetuning the model using 2019 dataset for training and a 2019 subset as CV.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "580550": "Hello there, I want to know if it is normal to mix the old and new dataset in a balanced way? are the distribution of the old and new data the same? Will the Model Generalize well if I do this Mixup? Thank you very much.",
    "582037": "mxcsyounes Rather than mixing the 2 datasets, which will make the task of making the cv troublesome and maybe less dependable. The following method shared by @drhabib has seemed to work, (https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100815#latest-582005) : Pretraining the model on 2015 dataset while using 2019 as validation, then finetuning the model using 2019 dataset for training and a 2019 subset as CV."
  },
  "source": "meta"
}