{
  "id": 227834,
  "title": "How to choose the correct threshold?",
  "url": "/competitions/vinbigdata-chest-xray-abnormalities-detection/discussion/227834",
  "author_name": "Jagadish Sivakumar",
  "post_date": "2021-03-22T13:22:53.090000",
  "votes": 5,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Kindly share your views on selecting the optimal threshold. I am not sure how to select the best threshold values. Is it just trial and error or any methods to find the best threshold?</p>",
  "messages": [
    {
      "id": 1248250,
      "postDate": "2021-03-22T13:22:53.090Z",
      "content": "<p>Kindly share your views on selecting the optimal threshold. I am not sure how to select the best threshold values. Is it just trial and error or any methods to find the best threshold?</p>",
      "rawMarkdown": "Kindly share your views on selecting the optimal threshold. I am not sure how to select the best threshold values. Is it just trial and error or any methods to find the best threshold?",
      "votes": 5
    },
    {
      "id": 1255652,
      "postDate": "2021-03-29T03:16:19.587Z",
      "content": "<p>The best way to tune any parameter is to use your local CV.</p>",
      "rawMarkdown": "The best way to tune any parameter is to use your local CV.",
      "votes": 6,
      "replies": [
        {
          "id": 1255737,
          "postDate": "2021-03-29T06:38:26.013Z",
          "content": "<p>To be fair, local CV is a bit tricky in a challenge with unknown filtering experts in the test set (they may or may not be different from train)</p>",
          "rawMarkdown": "To be fair, local CV is a bit tricky in a challenge with unknown filtering experts in the test set (they may or may not be different from train)",
          "votes": 3
        }
      ]
    },
    {
      "id": 1249149,
      "postDate": "2021-03-23T06:21:30.780Z",
      "content": "<p>Maybe threshold is the secret sause.. No comments about that anywhere 😅</p>",
      "rawMarkdown": "Maybe threshold is the secret sause.. No comments about that anywhere 😅",
      "votes": 2
    },
    {
      "id": 1255347,
      "postDate": "2021-03-28T17:47:43.110Z",
      "content": "<p>A single model needs a low threshold. But if there are different models of ensemble, you can choose a low threshold and a high threshold.</p>",
      "rawMarkdown": "A single model needs a low threshold. But if there are different models of ensemble, you can choose a low threshold and a high threshold."
    },
    {
      "id": 1249277,
      "postDate": "2021-03-23T08:09:10.387Z",
      "content": "<p>Which threshold?</p>",
      "rawMarkdown": "Which threshold?",
      "replies": [
        {
          "id": 1249346,
          "postDate": "2021-03-23T08:38:33.953Z",
          "content": "<p>The low and high threshold range?</p>\n<p><a href=\"https://www.kaggle.com/corochann/vinbigdata-2-class-classifier-complete-pipeline\" target=\"_blank\">https://www.kaggle.com/corochann/vinbigdata-2-class-classifier-complete-pipeline</a></p>",
          "rawMarkdown": "The low and high threshold range?\n \nhttps://www.kaggle.com/corochann/vinbigdata-2-class-classifier-complete-pipeline",
          "votes": 2
        },
        {
          "id": 1255228,
          "postDate": "2021-03-28T15:33:53.820Z",
          "content": "<p>Lol. That was a very specific threshold in a very specific hidden notebook.</p>\n<p>I spent a couple of days trying to calibrate that threshold. I couldn't.</p>",
          "rawMarkdown": "Lol. That was a very specific threshold in a very specific hidden notebook.\n\nI spent a couple of days trying to calibrate that threshold. I couldn't."
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1255652,
      "author_name": "Chris Deotte",
      "author_url": "",
      "post_date": "2021-03-29T03:16:19.587000",
      "content": "<p>The best way to tune any parameter is to use your local CV.</p>",
      "votes": 6,
      "replies": [
        {
          "id": 1255737,
          "author_name": "Pascal Pfeiffer",
          "author_url": "",
          "post_date": "2021-03-29T06:38:26.013000",
          "content": "<p>To be fair, local CV is a bit tricky in a challenge with unknown filtering experts in the test set (they may or may not be different from train)</p>",
          "votes": 3,
          "replies": []
        }
      ]
    },
    {
      "id": 1249149,
      "author_name": "Joseph",
      "author_url": "",
      "post_date": "2021-03-23T06:21:30.780000",
      "content": "<p>Maybe threshold is the secret sause.. No comments about that anywhere 😅</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1255347,
      "author_name": "Alien",
      "author_url": "",
      "post_date": "2021-03-28T17:47:43.110000",
      "content": "<p>A single model needs a low threshold. But if there are different models of ensemble, you can choose a low threshold and a high threshold.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1249277,
      "author_name": "Ahmet Erdem",
      "author_url": "",
      "post_date": "2021-03-23T08:09:10.387000",
      "content": "<p>Which threshold?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1249346,
          "author_name": "Jagadish Sivakumar",
          "author_url": "",
          "post_date": "2021-03-23T08:38:33.953000",
          "content": "<p>The low and high threshold range?</p>\n<p><a href=\"https://www.kaggle.com/corochann/vinbigdata-2-class-classifier-complete-pipeline\" target=\"_blank\">https://www.kaggle.com/corochann/vinbigdata-2-class-classifier-complete-pipeline</a></p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 1255228,
          "author_name": "Wilmer E. Henao",
          "author_url": "",
          "post_date": "2021-03-28T15:33:53.820000",
          "content": "<p>Lol. That was a very specific threshold in a very specific hidden notebook.</p>\n<p>I spent a couple of days trying to calibrate that threshold. I couldn't.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1248250": "Kindly share your views on selecting the optimal threshold. I am not sure how to select the best threshold values. Is it just trial and error or any methods to find the best threshold?",
    "1255652": "The best way to tune any parameter is to use your local CV.",
    "1249149": "Maybe threshold is the secret sause.. No comments about that anywhere 😅",
    "1255347": "A single model needs a low threshold. But if there are different models of ensemble, you can choose a low threshold and a high threshold.",
    "1249277": "Which threshold?"
  }
}