{
  "id": 451179,
  "title": "Are false positives punished?",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/451179",
  "author_name": "",
  "post_date": "2023-10-27T13:12:49.254953100Z",
  "votes": 15,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I'm using the following ground truth and submission dataframes. The final row in the submission is a False Positive. I get the same mAP value (0.9) with or without it. This shows that False Positives are not punished but the official explanation clearly tells that the False Positives are punished. Am I missing something here?</p>\n<p>Truth</p>\n<pre><code>        series_id   event      step\n  062cae666e2a   onset   \n  062cae666e2a  wakeup   \n  062cae666e2a   onset  \n</code></pre>\n<p>Submission</p>\n<pre><code>     row_id    step     series_id   event     score\n              062cae666e2a   onset  \n              062cae666e2a  wakeup  \n           062cae666e2a   onset  \n         08db4255286f   onset  \n</code></pre>",
  "messages": [
    {
      "id": "2501458",
      "postDate": "10/27/2023 13:12:49",
      "content": "<p>I'm using the following ground truth and submission dataframes. The final row in the submission is a False Positive. I get the same mAP value (0.9) with or without it. This shows that False Positives are not punished but the official explanation clearly tells that the False Positives are punished. Am I missing something here?</p>\n<p>Truth</p>\n<pre><code>        series_id   event      step\n  062cae666e2a   onset   \n  062cae666e2a  wakeup   \n  062cae666e2a   onset  \n</code></pre>\n<p>Submission</p>\n<pre><code>     row_id    step     series_id   event     score\n              062cae666e2a   onset  \n              062cae666e2a  wakeup  \n           062cae666e2a   onset  \n         08db4255286f   onset  \n</code></pre>",
      "rawMarkdown": "I'm using the following ground truth and submission dataframes. The final row in the submission is a False Positive. I get the same mAP value (0.9) with or without it. This shows that False Positives are not punished but the official explanation clearly tells that the False Positives are punished. Am I missing something here?\n\nTruth\n```python\n        series_id   event      step\n260  062cae666e2a   onset   22812.0\n261  062cae666e2a  wakeup   28656.0\n282  062cae666e2a   onset  211848.0\n```\n\nSubmission\n```python\n     row_id    step     series_id   event     score\n5         5   22836  062cae666e2a   onset  0.920281\n6         6   28668  062cae666e2a  wakeup  0.913717\n38       38  211836  062cae666e2a   onset  0.910108\n100     100  166236  08db4255286f   onset  0.920350\n```",
      "votes": null
    },
    {
      "id": "2501585",
      "postDate": "10/27/2023 14:35:37",
      "content": "<p>Perhaps it doesn't affect the score because the false positive's series has zero true events? </p>",
      "rawMarkdown": "Perhaps it doesn't affect the score because the false positive's series has zero true events?",
      "votes": null
    },
    {
      "id": "2501592",
      "postDate": "10/27/2023 14:44:42",
      "content": "<p>I added a True positive for 08db4255286f (close values in both submission and truth). Event Detection AP gives the same score with and without the False positive datapoint but of course different than the one above.</p>\n<p>Edit: Actually what I tried wasn't counted as a True Positive because it had a lower score than a False detection. So it was just discarded???</p>",
      "rawMarkdown": "I added a True positive for 08db4255286f (close values in both submission and truth). Event Detection AP gives the same score with and without the False positive datapoint but of course different than the one above.\n\nEdit: Actually what I tried wasn't counted as a True Positive because it had a lower score than a False detection. So it was just discarded???",
      "votes": null
    },
    {
      "id": "2501712",
      "postDate": "10/27/2023 16:06:09",
      "content": "<p>Had tried a couple of examples a few weeks ago and false positives had actually brought down AP score when calculated locally. 🤔 </p>",
      "rawMarkdown": "Had tried a couple of examples a few weeks ago and false positives had actually brought down AP score when calculated locally. 🤔",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2501585,
      "author_name": "quillen",
      "author_url": "",
      "post_date": "10/27/2023 14:35:37",
      "content": "<p>Perhaps it doesn't affect the score because the false positive's series has zero true events? </p>",
      "votes": null,
      "replies": [
        {
          "id": 2501592,
          "author_name": "tolgadincer",
          "author_url": "",
          "post_date": "10/27/2023 14:44:42",
          "content": "<p>I added a True positive for 08db4255286f (close values in both submission and truth). Event Detection AP gives the same score with and without the False positive datapoint but of course different than the one above.</p>\n<p>Edit: Actually what I tried wasn't counted as a True Positive because it had a lower score than a False detection. So it was just discarded???</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2501712,
      "author_name": "bkowshik",
      "author_url": "",
      "post_date": "10/27/2023 16:06:09",
      "content": "<p>Had tried a couple of examples a few weeks ago and false positives had actually brought down AP score when calculated locally. 🤔 </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2501458": "I'm using the following ground truth and submission dataframes. The final row in the submission is a False Positive. I get the same mAP value (0.9) with or without it. This shows that False Positives are not punished but the official explanation clearly tells that the False Positives are punished. Am I missing something here?\n\nTruth\n```python\n        series_id   event      step\n260  062cae666e2a   onset   22812.0\n261  062cae666e2a  wakeup   28656.0\n282  062cae666e2a   onset  211848.0\n```\n\nSubmission\n```python\n     row_id    step     series_id   event     score\n5         5   22836  062cae666e2a   onset  0.920281\n6         6   28668  062cae666e2a  wakeup  0.913717\n38       38  211836  062cae666e2a   onset  0.910108\n100     100  166236  08db4255286f   onset  0.920350\n```",
    "2501585": "Perhaps it doesn't affect the score because the false positive's series has zero true events?",
    "2501592": "I added a True positive for 08db4255286f (close values in both submission and truth). Event Detection AP gives the same score with and without the False positive datapoint but of course different than the one above.\n\nEdit: Actually what I tried wasn't counted as a True Positive because it had a lower score than a False detection. So it was just discarded???",
    "2501712": "Had tried a couple of examples a few weeks ago and false positives had actually brought down AP score when calculated locally. 🤔"
  },
  "source": "meta"
}