{
  "id": 236031,
  "title": "No threshold selection for 'healthy' class",
  "url": "/competitions/plant-pathology-2021-fgvc8/discussion/236031",
  "author_name": "",
  "post_date": "2021-05-02T13:24:31.738164700Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have seen in some submission notebooks that, there is no threshold selected for healthy class in this competition, and those submission scores good even. Can someone please tell me the reason for doing it, healthy is even a class too, why not selecting a threshold value for it.</p>",
  "messages": [
    {
      "id": "1290844",
      "postDate": "05/02/2021 13:24:31",
      "content": "<p>I have seen in some submission notebooks that, there is no threshold selected for healthy class in this competition, and those submission scores good even. Can someone please tell me the reason for doing it, healthy is even a class too, why not selecting a threshold value for it.</p>",
      "rawMarkdown": "I have seen in some submission notebooks that, there is no threshold selected for healthy class in this competition, and those submission scores good even. Can someone please tell me the reason for doing it, healthy is even a class too, why not selecting a threshold value for it.",
      "votes": null
    },
    {
      "id": "1291052",
      "postDate": "05/02/2021 16:59:19",
      "content": "<p>Well, \"healthy\" is a class that excludes other classes. In principle, you can assign \"healthy\" class to the sample if predicted probability of other classes is lower than threshold. But I think it's better to set threshold for \"healthy\" as well.</p>",
      "rawMarkdown": "Well, \"healthy\" is a class that excludes other classes. In principle, you can assign \"healthy\" class to the sample if predicted probability of other classes is lower than threshold. But I think it's better to set threshold for \"healthy\" as well.",
      "votes": null
    },
    {
      "id": "1291144",
      "postDate": "05/02/2021 18:09:42",
      "content": "<p>Because they is no plant healthy and have a disease in the same time. The threshold is a hyperplane that separate between healthy and not. Also they are many ways to calculate the threshold the best one is to don't choose it manually and generate a classification model take as input 5 variable (the output of the first model) and give as input the class of categories (12 classes)</p>",
      "rawMarkdown": "Because they is no plant healthy and have a disease in the same time. The threshold is a hyperplane that separate between healthy and not. Also they are many ways to calculate the threshold the best one is to don't choose it manually and generate a classification model take as input 5 variable (the output of the first model) and give as input the class of categories (12 classes)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1291052,
      "author_name": "atamazian",
      "author_url": "",
      "post_date": "05/02/2021 16:59:19",
      "content": "<p>Well, \"healthy\" is a class that excludes other classes. In principle, you can assign \"healthy\" class to the sample if predicted probability of other classes is lower than threshold. But I think it's better to set threshold for \"healthy\" as well.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1291144,
      "author_name": "aithammadiabdellatif",
      "author_url": "",
      "post_date": "05/02/2021 18:09:42",
      "content": "<p>Because they is no plant healthy and have a disease in the same time. The threshold is a hyperplane that separate between healthy and not. Also they are many ways to calculate the threshold the best one is to don't choose it manually and generate a classification model take as input 5 variable (the output of the first model) and give as input the class of categories (12 classes)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1290844": "I have seen in some submission notebooks that, there is no threshold selected for healthy class in this competition, and those submission scores good even. Can someone please tell me the reason for doing it, healthy is even a class too, why not selecting a threshold value for it.",
    "1291052": "Well, \"healthy\" is a class that excludes other classes. In principle, you can assign \"healthy\" class to the sample if predicted probability of other classes is lower than threshold. But I think it's better to set threshold for \"healthy\" as well.",
    "1291144": "Because they is no plant healthy and have a disease in the same time. The threshold is a hyperplane that separate between healthy and not. Also they are many ways to calculate the threshold the best one is to don't choose it manually and generate a classification model take as input 5 variable (the output of the first model) and give as input the class of categories (12 classes)"
  },
  "source": "meta"
}