{
  "id": 221382,
  "title": "Best strategy to create folds from tfrec.?",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/221382",
  "author_name": "",
  "post_date": "2021-02-22T14:24:30.090481500Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>How do you separate the tfrec data for training and validation?</p>",
  "messages": [
    {
      "id": "1214002",
      "postDate": "02/22/2021 14:24:30",
      "content": "<p>How do you separate the tfrec data for training and validation?</p>",
      "rawMarkdown": "How do you separate the tfrec data for training and validation?",
      "votes": null
    },
    {
      "id": "1214027",
      "postDate": "02/22/2021 14:43:20",
      "content": "<p>Hello!</p>\n<p>Working with TFRecords provided by organizers can be a pain because you're unable to separate those files into folds with stratification and respect to <code>PatientID</code> (which is important in this competition). I suggest using TFRecords made by other users, e.g. this <strong><a href=\"https://www.kaggle.com/nickuzmenkov/ranzcr-clip-kfold-tfrecords\" target=\"_blank\">dataset</a></strong> of 600x600 image quality or a bunch of lower image quality datasets <strong><a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/208689\" target=\"_blank\">here</a></strong>. </p>\n<p>All TFRecords in these datasets are organized so that label-wise distributions are held similar to the original distribution and no patient shares multiple folds.</p>",
      "rawMarkdown": "Hello!\n\nWorking with TFRecords provided by organizers can be a pain because you're unable to separate those files into folds with stratification and respect to `PatientID` (which is important in this competition). I suggest using TFRecords made by other users, e.g. this **[dataset](https://www.kaggle.com/nickuzmenkov/ranzcr-clip-kfold-tfrecords)** of 600x600 image quality or a bunch of lower image quality datasets **[here](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/208689)**. \n\nAll TFRecords in these datasets are organized so that label-wise distributions are held similar to the original distribution and no patient shares multiple folds.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1214027,
      "author_name": "nickuzmenkov",
      "author_url": "",
      "post_date": "02/22/2021 14:43:20",
      "content": "<p>Hello!</p>\n<p>Working with TFRecords provided by organizers can be a pain because you're unable to separate those files into folds with stratification and respect to <code>PatientID</code> (which is important in this competition). I suggest using TFRecords made by other users, e.g. this <strong><a href=\"https://www.kaggle.com/nickuzmenkov/ranzcr-clip-kfold-tfrecords\" target=\"_blank\">dataset</a></strong> of 600x600 image quality or a bunch of lower image quality datasets <strong><a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/208689\" target=\"_blank\">here</a></strong>. </p>\n<p>All TFRecords in these datasets are organized so that label-wise distributions are held similar to the original distribution and no patient shares multiple folds.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1214002": "How do you separate the tfrec data for training and validation?",
    "1214027": "Hello!\n\nWorking with TFRecords provided by organizers can be a pain because you're unable to separate those files into folds with stratification and respect to `PatientID` (which is important in this competition). I suggest using TFRecords made by other users, e.g. this **[dataset](https://www.kaggle.com/nickuzmenkov/ranzcr-clip-kfold-tfrecords)** of 600x600 image quality or a bunch of lower image quality datasets **[here](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/208689)**. \n\nAll TFRecords in these datasets are organized so that label-wise distributions are held similar to the original distribution and no patient shares multiple folds."
  },
  "source": "meta"
}