{
  "id": 189997,
  "title": "Use 0 as a prediction",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/189997",
  "author_name": "TOUCAN",
  "post_date": "2020-10-09T16:21:56.021000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi, </p>\n<p>I would like to know if this is possible to put 0 (or 1) as a prediction without getting an error (with the calculation of the error with loss). </p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": 1044388,
      "postDate": "2020-10-09T17:58:17.223Z",
      "content": "<p>As with the last RSNA competition, the host most likely put a small epsilon as floor and 1-epsilon as the ceiling of your predictions to prevent the grader throwing an error. I'm actually surprised they have not explicitly mention that anywhere in the competition rules.</p>\n<p>see previous competition evaluation:<br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/evaluation</a></p>",
      "rawMarkdown": "As with the last RSNA competition, the host most likely put a small epsilon as floor and 1-epsilon as the ceiling of your predictions to prevent the grader throwing an error. I'm actually surprised they have not explicitly mention that anywhere in the competition rules.\n\nsee previous competition evaluation:\n[https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/evaluation](url)",
      "votes": 2,
      "replies": [
        {
          "id": 1044419,
          "postDate": "2020-10-09T18:39:15.397Z",
          "content": "<p>Thank you, I am struggling with \"Submission Scoring Error\" and I thought it could be this. I will change this part of my code to see if it solves the issue.</p>",
          "rawMarkdown": "Thank you, I am struggling with \"Submission Scoring Error\" and I thought it could be this. I will change this part of my code to see if it solves the issue."
        },
        {
          "id": 1044629,
          "postDate": "2020-10-10T01:51:38.433Z",
          "content": "<p>I agree with Yee. Kaggle must be adding an epsilon. In last year's competition, they added <code>1e-15</code>. So all of our submitted predictions most likely get redefined as <code>p = max(min(p,1−1e−15),1e−15) .</code></p>",
          "rawMarkdown": "I agree with Yee. Kaggle must be adding an epsilon. In last year's competition, they added `1e-15`. So all of our submitted predictions most likely get redefined as `p = max(min(p,1−1e−15),1e−15) .`"
        }
      ]
    },
    {
      "id": 1044251,
      "postDate": "2020-10-09T16:21:56.020Z",
      "content": "<p>Hi, </p>\n<p>I would like to know if this is possible to put 0 (or 1) as a prediction without getting an error (with the calculation of the error with loss). </p>\n<p>Thanks</p>",
      "rawMarkdown": "Hi, \n\nI would like to know if this is possible to put 0 (or 1) as a prediction without getting an error (with the calculation of the error with loss). \n\nThanks"
    }
  ],
  "comments": [
    {
      "id": 1044388,
      "author_name": "Yee Ng",
      "author_url": "",
      "post_date": "2020-10-09T17:58:17.223000",
      "content": "<p>As with the last RSNA competition, the host most likely put a small epsilon as floor and 1-epsilon as the ceiling of your predictions to prevent the grader throwing an error. I'm actually surprised they have not explicitly mention that anywhere in the competition rules.</p>\n<p>see previous competition evaluation:<br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/evaluation</a></p>",
      "votes": 2,
      "replies": [
        {
          "id": 1044419,
          "author_name": "TOUCAN",
          "author_url": "",
          "post_date": "2020-10-09T18:39:15.397000",
          "content": "<p>Thank you, I am struggling with \"Submission Scoring Error\" and I thought it could be this. I will change this part of my code to see if it solves the issue.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1044629,
          "author_name": "Chris Deotte",
          "author_url": "",
          "post_date": "2020-10-10T01:51:38.433000",
          "content": "<p>I agree with Yee. Kaggle must be adding an epsilon. In last year's competition, they added <code>1e-15</code>. So all of our submitted predictions most likely get redefined as <code>p = max(min(p,1−1e−15),1e−15) .</code></p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1044388": "As with the last RSNA competition, the host most likely put a small epsilon as floor and 1-epsilon as the ceiling of your predictions to prevent the grader throwing an error. I'm actually surprised they have not explicitly mention that anywhere in the competition rules.\n\nsee previous competition evaluation:\n[https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/evaluation](url)",
    "1044251": "Hi, \n\nI would like to know if this is possible to put 0 (or 1) as a prediction without getting an error (with the calculation of the error with loss). \n\nThanks"
  }
}