{
  "id": 180506,
  "title": "Splitted Loss",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/180506",
  "author_name": "Dmitrij Kozachuk",
  "post_date": "2020-09-05T09:48:28",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>One of the challanges of this competition is an unusual metric. We have to return regression prediction and it's confidence simultaniously, but usually we're asked to return either class prediction with probability or only regression one. I glad to present you a notebook with new loss for this competition named <a href=\"https://www.kaggle.com/koza4ukdmitrij/splitted-loss\" target=\"_blank\">Splitted Loss</a>. This notebook presents a method for solving this task with Splitted Loss: task will be splitted into two part and for each part particular loss will be presented with high correlation with target Laplace Loss. Mathematical proof for Splitted Loss will be given with corresponding implementations and pictures for some cases. Moreover, some insights under Laplace Loss will be covered that can helps in this competition even without Splitted Loss using. I don't use this in this competition, so please tell how it affects your metric if you've decided to use it.</p>",
  "messages": [
    {
      "id": 999050,
      "postDate": "2020-09-05T09:48:28Z",
      "content": "<p>One of the challanges of this competition is an unusual metric. We have to return regression prediction and it's confidence simultaniously, but usually we're asked to return either class prediction with probability or only regression one. I glad to present you a notebook with new loss for this competition named <a href=\"https://www.kaggle.com/koza4ukdmitrij/splitted-loss\" target=\"_blank\">Splitted Loss</a>. This notebook presents a method for solving this task with Splitted Loss: task will be splitted into two part and for each part particular loss will be presented with high correlation with target Laplace Loss. Mathematical proof for Splitted Loss will be given with corresponding implementations and pictures for some cases. Moreover, some insights under Laplace Loss will be covered that can helps in this competition even without Splitted Loss using. I don't use this in this competition, so please tell how it affects your metric if you've decided to use it.</p>",
      "rawMarkdown": "   One of the challanges of this competition is an unusual metric. We have to return regression prediction and it's confidence simultaniously, but usually we're asked to return either class prediction with probability or only regression one. I glad to present you a notebook with new loss for this competition named [Splitted Loss](https://www.kaggle.com/koza4ukdmitrij/splitted-loss). This notebook presents a method for solving this task with Splitted Loss: task will be splitted into two part and for each part particular loss will be presented with high correlation with target Laplace Loss. Mathematical proof for Splitted Loss will be given with corresponding implementations and pictures for some cases. Moreover, some insights under Laplace Loss will be covered that can helps in this competition even without Splitted Loss using. I don't use this in this competition, so please tell how it affects your metric if you've decided to use it."
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "999050": "   One of the challanges of this competition is an unusual metric. We have to return regression prediction and it's confidence simultaniously, but usually we're asked to return either class prediction with probability or only regression one. I glad to present you a notebook with new loss for this competition named [Splitted Loss](https://www.kaggle.com/koza4ukdmitrij/splitted-loss). This notebook presents a method for solving this task with Splitted Loss: task will be splitted into two part and for each part particular loss will be presented with high correlation with target Laplace Loss. Mathematical proof for Splitted Loss will be given with corresponding implementations and pictures for some cases. Moreover, some insights under Laplace Loss will be covered that can helps in this competition even without Splitted Loss using. I don't use this in this competition, so please tell how it affects your metric if you've decided to use it."
  }
}