{
  "id": 497994,
  "title": "Do you think the total probability of each bird species in the SUBMISSION file is 100%?",
  "url": "/competitions/birdclef-2024/discussion/497994",
  "author_name": "",
  "post_date": "2024-04-26T14:33:16.950417500Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Do you think the total probability of each bird species in the SUBMISSION file is 100%?<br>\nFor example, if more than one bird species is listed in secondary_labels, I think the highest goal is for both primary_labels and secondary_labels to be 100%, but is this concept different?</p>",
  "messages": [
    {
      "id": "2777131",
      "postDate": "04/26/2024 14:33:16",
      "content": "<p>Do you think the total probability of each bird species in the SUBMISSION file is 100%?<br>\nFor example, if more than one bird species is listed in secondary_labels, I think the highest goal is for both primary_labels and secondary_labels to be 100%, but is this concept different?</p>",
      "rawMarkdown": "Do you think the total probability of each bird species in the SUBMISSION file is 100%?\nFor example, if more than one bird species is listed in secondary_labels, I think the highest goal is for both primary_labels and secondary_labels to be 100%, but is this concept different?",
      "votes": null
    },
    {
      "id": "2777404",
      "postDate": "04/26/2024 16:23:22",
      "content": "<p>No.</p>\n<p>This is a multi label problem, not a multi class problem. Metric is computed species by species. If you multiply all predictions of one species by a constant, metric does not change. Same if you add a constant to all the predictions for a given species.</p>\n<p>Normalizing to have all predictions sum to 1 row by row is not required and probably is detrimental to your score.</p>",
      "rawMarkdown": "No.\n\nThis is a multi label problem, not a multi class problem. Metric is computed species by species. If you multiply all predictions of one species by a constant, metric does not change. Same if you add a constant to all the predictions for a given species.\n\nNormalizing to have all predictions sum to 1 row by row is not required and probably is detrimental to your score.",
      "votes": null
    },
    {
      "id": "2795029",
      "postDate": "05/05/2024 16:03:11",
      "content": "<p>No.</p>\n<p>The sum of all probabilities for each row must be 100% as you are predicting the probability of a species being present in that specific case/row. However, the sum of all probabilities for a certain bird species (class) is not necessarily 100% as you are not guaranteed in any row that a specific class is present. You can have 1s for all rows in a specific column(species). In this case all the other probabilities should be 0.</p>\n<p>Please, correct me if I am wrong.</p>",
      "rawMarkdown": "No.\n\nThe sum of all probabilities for each row must be 100% as you are predicting the probability of a species being present in that specific case/row. However, the sum of all probabilities for a certain bird species (class) is not necessarily 100% as you are not guaranteed in any row that a specific class is present. You can have 1s for all rows in a specific column(species). In this case all the other probabilities should be 0.\n\nPlease, correct me if I am wrong.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2777404,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "04/26/2024 16:23:22",
      "content": "<p>No.</p>\n<p>This is a multi label problem, not a multi class problem. Metric is computed species by species. If you multiply all predictions of one species by a constant, metric does not change. Same if you add a constant to all the predictions for a given species.</p>\n<p>Normalizing to have all predictions sum to 1 row by row is not required and probably is detrimental to your score.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2795029,
      "author_name": "davitkhalatyan3",
      "author_url": "",
      "post_date": "05/05/2024 16:03:11",
      "content": "<p>No.</p>\n<p>The sum of all probabilities for each row must be 100% as you are predicting the probability of a species being present in that specific case/row. However, the sum of all probabilities for a certain bird species (class) is not necessarily 100% as you are not guaranteed in any row that a specific class is present. You can have 1s for all rows in a specific column(species). In this case all the other probabilities should be 0.</p>\n<p>Please, correct me if I am wrong.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2777131": "Do you think the total probability of each bird species in the SUBMISSION file is 100%?\nFor example, if more than one bird species is listed in secondary_labels, I think the highest goal is for both primary_labels and secondary_labels to be 100%, but is this concept different?",
    "2777404": "No.\n\nThis is a multi label problem, not a multi class problem. Metric is computed species by species. If you multiply all predictions of one species by a constant, metric does not change. Same if you add a constant to all the predictions for a given species.\n\nNormalizing to have all predictions sum to 1 row by row is not required and probably is detrimental to your score.",
    "2795029": "No.\n\nThe sum of all probabilities for each row must be 100% as you are predicting the probability of a species being present in that specific case/row. However, the sum of all probabilities for a certain bird species (class) is not necessarily 100% as you are not guaranteed in any row that a specific class is present. You can have 1s for all rows in a specific column(species). In this case all the other probabilities should be 0.\n\nPlease, correct me if I am wrong."
  },
  "source": "meta"
}