{
  "competition": "rsna-knee-abnormality-detection",
  "topic_id": "742327",
  "comments": [
    {
      "id": 3526643,
      "authorName": "Prateek Grover",
      "votes": 1,
      "postDate": "2026-09-21T22:13:36.547000",
      "content": "<p>It's hard to say, but Top 50, everyone is now over 0.95 - so I believe they all do indeed have separate models outperforming the public stack's 0.941 number. (atleast I do). </p>\n<p>Perhaps in a couple of weeks, when teams start forming ensembles - we'll see atleast the entire Top 10 go over 0.96, and they should remain there with the Private LB too.</p>"
    },
    {
      "id": 3526708,
      "authorName": "Raymond Yuen",
      "votes": 0,
      "postDate": "2026-09-22T04:15:24.810000",
      "content": "<p>Hopefully we’re not overfitting, and hopefully we won’t be🫣. It’s just that with this structure, it feels hard to actually tell whether a model is genuinely better or just happens to line up with the public LB.</p>"
    },
    {
      "id": 3526524,
      "authorName": "PC Jimmmy",
      "votes": 1,
      "postDate": "2026-09-21T16:54:08.477000",
      "content": "<p>I have not looked close but I think most of the pile up in bronze and some silver are two shared notebooks that have been tweaked or just copied by lots of folks.  When a huge portion of the metals are based on only a couple of shared notebooks the potential for shakeup is huge.  I once jumped up 1000 positions into a silver in a competition where most of the pre-final metals were the result of one shared notebook that \"failed\".</p>"
    },
    {
      "id": 3526707,
      "authorName": "Raymond Yuen",
      "votes": 0,
      "postDate": "2026-09-22T04:13:20.777000",
      "content": "<p>wow. a 1000 places jump is insane! I guess the biggest question is whether the shared solutions are robust or people are just sitting on the same lb overfit</p>"
    }
  ],
  "messages": [],
  "raw_show": {
    "topic": {
      "id": 742327,
      "title": "How bad will the private LB shake-up be?",
      "authorName": "Raymond Yuen",
      "commentCount": 4,
      "votes": 3,
      "postDate": "2026-09-21T13:46:50.778000"
    },
    "comments": [
      {
        "id": 3526643,
        "authorName": "Prateek Grover",
        "votes": 1,
        "postDate": "2026-09-21T22:13:36.547000",
        "content": "<p>It's hard to say, but Top 50, everyone is now over 0.95 - so I believe they all do indeed have separate models outperforming the public stack's 0.941 number. (atleast I do). </p>\n<p>Perhaps in a couple of weeks, when teams start forming ensembles - we'll see atleast the entire Top 10 go over 0.96, and they should remain there with the Private LB too.</p>"
      },
      {
        "id": 3526708,
        "authorName": "Raymond Yuen",
        "votes": 0,
        "postDate": "2026-09-22T04:15:24.810000",
        "content": "<p>Hopefully we’re not overfitting, and hopefully we won’t be🫣. It’s just that with this structure, it feels hard to actually tell whether a model is genuinely better or just happens to line up with the public LB.</p>"
      },
      {
        "id": 3526524,
        "authorName": "PC Jimmmy",
        "votes": 1,
        "postDate": "2026-09-21T16:54:08.477000",
        "content": "<p>I have not looked close but I think most of the pile up in bronze and some silver are two shared notebooks that have been tweaked or just copied by lots of folks.  When a huge portion of the metals are based on only a couple of shared notebooks the potential for shakeup is huge.  I once jumped up 1000 positions into a silver in a competition where most of the pre-final metals were the result of one shared notebook that \"failed\".</p>"
      },
      {
        "id": 3526707,
        "authorName": "Raymond Yuen",
        "votes": 0,
        "postDate": "2026-09-22T04:13:20.777000",
        "content": "<p>wow. a 1000 places jump is insane! I guess the biggest question is whether the shared solutions are robust or people are just sitting on the same lb overfit</p>"
      }
    ]
  },
  "topic": {
    "id": 742327,
    "title": "How bad will the private LB shake-up be?",
    "authorName": "Raymond Yuen",
    "commentCount": 4,
    "votes": 3,
    "postDate": "2026-09-21T13:46:50.778000"
  },
  "index": {
    "id": "742327",
    "title": "How bad will the private LB shake-up be?",
    "authorName": "",
    "commentCount": "4",
    "votes": "3",
    "postDate": "2026-09-21 13:46:50.778000"
  }
}