{
  "id": 166320,
  "title": "How does different sampling techniques makes difference?",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/166320",
  "author_name": "sulabh tiwari",
  "post_date": "2020-07-12T13:42:18.924000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello Guys,\nI am trying to understand the effect of sampling technique that I have implemented. Basically to fight with class imbalance i broke big dataset into (4 [class 0] : 1 [class 1]). This has resulted in to ~16 dataset. On every dataset I have applied weighted random sampler technique. The result was ok for me as the training was very fast in very few epochs i reached the good score. I have made LB of 0.925 on 384 dataset. I didnt used tabular data. no external data. Moreover I still see the posibility to increase it further as I have not used efficientnet and optimized my local CV. This looks fair to me.</p>\n\n<p>But does it really makes sense to deal with this proble</p>",
  "messages": [
    {
      "id": 926089,
      "postDate": "2020-07-12T13:42:18.923Z",
      "content": "<p>Hello Guys,\nI am trying to understand the effect of sampling technique that I have implemented. Basically to fight with class imbalance i broke big dataset into (4 [class 0] : 1 [class 1]). This has resulted in to ~16 dataset. On every dataset I have applied weighted random sampler technique. The result was ok for me as the training was very fast in very few epochs i reached the good score. I have made LB of 0.925 on 384 dataset. I didnt used tabular data. no external data. Moreover I still see the posibility to increase it further as I have not used efficientnet and optimized my local CV. This looks fair to me.</p>\n\n<p>But does it really makes sense to deal with this proble</p>",
      "rawMarkdown": "Hello Guys,\nI am trying to understand the effect of sampling technique that I have implemented. Basically to fight with class imbalance i broke big dataset into (4 [class 0] : 1 [class 1]). This has resulted in to ~16 dataset. On every dataset I have applied weighted random sampler technique. The result was ok for me as the training was very fast in very few epochs i reached the good score. I have made LB of 0.925 on 384 dataset. I didnt used tabular data. no external data. Moreover I still see the posibility to increase it further as I have not used efficientnet and optimized my local CV. This looks fair to me.\n\nBut does it really makes sense to deal with this proble"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "926089": "Hello Guys,\nI am trying to understand the effect of sampling technique that I have implemented. Basically to fight with class imbalance i broke big dataset into (4 [class 0] : 1 [class 1]). This has resulted in to ~16 dataset. On every dataset I have applied weighted random sampler technique. The result was ok for me as the training was very fast in very few epochs i reached the good score. I have made LB of 0.925 on 384 dataset. I didnt used tabular data. no external data. Moreover I still see the posibility to increase it further as I have not used efficientnet and optimized my local CV. This looks fair to me.\n\nBut does it really makes sense to deal with this proble"
  }
}