{
  "id": 191913,
  "title": "[For LB leaders] How much model training for top LB scores?",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/191913",
  "author_name": "imnishantg",
  "post_date": "2020-10-19T11:08:17.370000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi Everyone</p>\n<p>To test out my hypothesis on model structure, data augmentations, etc; I'm training my model on 500 studies (not images) for one epoch to see how is it comparing.<br>\nI'm usually getting 0.8 - 1.1 on competition eval metric with that much training. Possibly, I'm training too less.<br>\nSince I wanted to save the GPU time to experiment broadly, I was limiting training time. </p>\n<p>Wanted to know from the LB leaders, how much model training are you doing (#studies or images, #epochs, etc.) to get to 0.15 - 0.3 scores?</p>\n<p>Thanks in advance.</p>",
  "messages": [
    {
      "id": 1053784,
      "postDate": "2020-10-19T11:08:17.370Z",
      "content": "<p>Hi Everyone</p>\n<p>To test out my hypothesis on model structure, data augmentations, etc; I'm training my model on 500 studies (not images) for one epoch to see how is it comparing.<br>\nI'm usually getting 0.8 - 1.1 on competition eval metric with that much training. Possibly, I'm training too less.<br>\nSince I wanted to save the GPU time to experiment broadly, I was limiting training time. </p>\n<p>Wanted to know from the LB leaders, how much model training are you doing (#studies or images, #epochs, etc.) to get to 0.15 - 0.3 scores?</p>\n<p>Thanks in advance.</p>",
      "rawMarkdown": "Hi Everyone\n\nTo test out my hypothesis on model structure, data augmentations, etc; I'm training my model on 500 studies (not images) for one epoch to see how is it comparing.\nI'm usually getting 0.8 - 1.1 on competition eval metric with that much training. Possibly, I'm training too less.\nSince I wanted to save the GPU time to experiment broadly, I was limiting training time. \n\nWanted to know from the LB leaders, how much model training are you doing (#studies or images, #epochs, etc.) to get to 0.15 - 0.3 scores?\n\nThanks in advance.\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1053784": "Hi Everyone\n\nTo test out my hypothesis on model structure, data augmentations, etc; I'm training my model on 500 studies (not images) for one epoch to see how is it comparing.\nI'm usually getting 0.8 - 1.1 on competition eval metric with that much training. Possibly, I'm training too less.\nSince I wanted to save the GPU time to experiment broadly, I was limiting training time. \n\nWanted to know from the LB leaders, how much model training are you doing (#studies or images, #epochs, etc.) to get to 0.15 - 0.3 scores?\n\nThanks in advance.\n"
  }
}