{
  "id": 424036,
  "title": "Which precision and recall for class 0 do you get?",
  "url": "/competitions/anomaly-class-detection/discussion/424036",
  "author_name": "",
  "post_date": "2023-07-12T09:45:34.553162800Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>The main metric of the competition is the F1-Score calculated for both classes. But for me, as the host of the competition, it is very important to know/to get best precision and recall combination for minority-anomaly class.</p>\n<p>For example, I wish to have precision 90% and recall 30%, if it is possible or any other, but with high value for precision and not very low value for recall.</p>\n<p>So, dear participants, please, share your precision and recall results for class 0.</p>",
  "messages": [
    {
      "id": "2341796",
      "postDate": "07/12/2023 09:45:34",
      "content": "<p>The main metric of the competition is the F1-Score calculated for both classes. But for me, as the host of the competition, it is very important to know/to get best precision and recall combination for minority-anomaly class.</p>\n<p>For example, I wish to have precision 90% and recall 30%, if it is possible or any other, but with high value for precision and not very low value for recall.</p>\n<p>So, dear participants, please, share your precision and recall results for class 0.</p>",
      "rawMarkdown": "The main metric of the competition is the F1-Score calculated for both classes. But for me, as the host of the competition, it is very important to know/to get best precision and recall combination for minority-anomaly class.\n\nFor example, I wish to have precision 90% and recall 30%, if it is possible or any other, but with high value for precision and not very low value for recall.\n\nSo, dear participants, please, share your precision and recall results for class 0.",
      "votes": null
    },
    {
      "id": "2341798",
      "postDate": "07/12/2023 09:47:00",
      "content": "<p><a href=\"https://www.kaggle.com/lucasmorin\" target=\"_blank\">@lucasmorin</a> Welcome to share your thoughts! Is it possible to achieve such results with this dataset?</p>",
      "rawMarkdown": "lucasmorin Welcome to share your thoughts! Is it possible to achieve such results with this dataset?",
      "votes": null
    },
    {
      "id": "2361406",
      "postDate": "07/27/2023 11:46:45",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/kindspidey787\" target=\"_blank\">@kindspidey787</a> I am afraid the PR curve for 0 is a bit ugly. I've updated my logistic regression notebook to include more details about the results… This is not the best model, but it seems to be a strong baseline. </p>\n<p>You can get high precision, but as expected the recall would be abysmal… the best micro F1 is actually with a higher threshold (somehow balanced precision and recall). </p>\n<p>Further discussion would be difficult without factoring the returns. </p>",
      "rawMarkdown": "Hi @kindspidey787 I am afraid the PR curve for 0 is a bit ugly. I've updated my logistic regression notebook to include more details about the results... This is not the best model, but it seems to be a strong baseline. \n\nYou can get high precision, but as expected the recall would be abysmal... the best micro F1 is actually with a higher threshold (somehow balanced precision and recall). \n\nFurther discussion would be difficult without factoring the returns.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2341798,
      "author_name": "",
      "author_url": "",
      "post_date": "07/12/2023 09:47:00",
      "content": "<p><a href=\"https://www.kaggle.com/lucasmorin\" target=\"_blank\">@lucasmorin</a> Welcome to share your thoughts! Is it possible to achieve such results with this dataset?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2361406,
          "author_name": "lucasmorin",
          "author_url": "",
          "post_date": "07/27/2023 11:46:45",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/kindspidey787\" target=\"_blank\">@kindspidey787</a> I am afraid the PR curve for 0 is a bit ugly. I've updated my logistic regression notebook to include more details about the results… This is not the best model, but it seems to be a strong baseline. </p>\n<p>You can get high precision, but as expected the recall would be abysmal… the best micro F1 is actually with a higher threshold (somehow balanced precision and recall). </p>\n<p>Further discussion would be difficult without factoring the returns. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2341796": "The main metric of the competition is the F1-Score calculated for both classes. But for me, as the host of the competition, it is very important to know/to get best precision and recall combination for minority-anomaly class.\n\nFor example, I wish to have precision 90% and recall 30%, if it is possible or any other, but with high value for precision and not very low value for recall.\n\nSo, dear participants, please, share your precision and recall results for class 0.",
    "2341798": "lucasmorin Welcome to share your thoughts! Is it possible to achieve such results with this dataset?",
    "2361406": "Hi @kindspidey787 I am afraid the PR curve for 0 is a bit ugly. I've updated my logistic regression notebook to include more details about the results... This is not the best model, but it seems to be a strong baseline. \n\nYou can get high precision, but as expected the recall would be abysmal... the best micro F1 is actually with a higher threshold (somehow balanced precision and recall). \n\nFurther discussion would be difficult without factoring the returns."
  },
  "source": "meta"
}