{
  "id": 110177,
  "title": "Leak. Wait when solved!",
  "url": "/competitions/3d-object-detection-for-autonomous-vehicles/discussion/110177",
  "author_name": "Mukharbek Organokov",
  "post_date": "2019-09-25T15:28:54.938000",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Guys,</p>\n\n<p>There is a leak in data or something wrong happened, that's why let's just wait when it is solved. \nJust do not waste of time forking, submitting 0.577 kernels.</p>\n\n<p>As you have seen, there are 0.577 dummy score with this kernel\n<a href=\"https://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577\">https://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577</a></p>\n\n<p>Which is obviously just a forked of the original one \n<a href=\"https://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel\">https://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel</a></p>\n\n<p>with no changes, I think. </p>\n\n<p>But you can take the original one with 0.231 and changing variables like put something weird ecpochs=25 and get same 0.577. Even higher than in forked version ))) </p>\n\n<p>What's the difference between all this 0.577 I don't know, but we played around and managed to get different improvements of \"0.577\"  value)))</p>\n\n<p>So, let's come back when Lyft will find what's going on )))</p>",
  "messages": [
    {
      "id": 633937,
      "postDate": "2019-09-25T15:55:17.410Z",
      "content": "<p>Hi!  Yes, saw this in <a href=\"/joonl04\">@joonl04</a>'s kernel yesterday, we've been working on it. It's not a leak, it's an error in the metric. We'll finish fixing it, rescore, and put up a thread on the forums. Thanks!</p>",
      "rawMarkdown": "Hi!  Yes, saw this in @joonl04's kernel yesterday, we've been working on it. It's not a leak, it's an error in the metric. We'll finish fixing it, rescore, and put up a thread on the forums. Thanks!",
      "votes": 4,
      "replies": [
        {
          "id": 634006,
          "postDate": "2019-09-25T17:33:17.993Z",
          "content": "<p>Great to hear it's not a leak. Looking forward to getting back into this comp. </p>",
          "rawMarkdown": "Great to hear it's not a leak. Looking forward to getting back into this comp. "
        }
      ]
    },
    {
      "id": 633923,
      "postDate": "2019-09-25T15:28:54.937Z",
      "content": "<p>Guys,</p>\n\n<p>There is a leak in data or something wrong happened, that's why let's just wait when it is solved. \nJust do not waste of time forking, submitting 0.577 kernels.</p>\n\n<p>As you have seen, there are 0.577 dummy score with this kernel\n<a href=\"https://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577\">https://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577</a></p>\n\n<p>Which is obviously just a forked of the original one \n<a href=\"https://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel\">https://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel</a></p>\n\n<p>with no changes, I think. </p>\n\n<p>But you can take the original one with 0.231 and changing variables like put something weird ecpochs=25 and get same 0.577. Even higher than in forked version ))) </p>\n\n<p>What's the difference between all this 0.577 I don't know, but we played around and managed to get different improvements of \"0.577\"  value)))</p>\n\n<p>So, let's come back when Lyft will find what's going on )))</p>",
      "rawMarkdown": "Guys,\n\nThere is a leak in data or something wrong happened, that's why let's just wait when it is solved. \nJust do not waste of time forking, submitting 0.577 kernels.\n\nAs you have seen, there are 0.577 dummy score with this kernel\nhttps://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577\n\nWhich is obviously just a forked of the original one \nhttps://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel\n\nwith no changes, I think. \n\nBut you can take the original one with 0.231 and changing variables like put something weird ecpochs=25 and get same 0.577. Even higher than in forked version ))) \n\nWhat's the difference between all this 0.577 I don't know, but we played around and managed to get different improvements of \"0.577\"  value)))\n\nSo, let's come back when Lyft will find what's going on )))\n\n ",
      "votes": 3
    },
    {
      "id": 633950,
      "postDate": "2019-09-25T16:01:34.133Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 633937,
      "author_name": "Phil Culliton",
      "author_url": "",
      "post_date": "2019-09-25T15:55:17.410000",
      "content": "<p>Hi!  Yes, saw this in <a href=\"/joonl04\">@joonl04</a>'s kernel yesterday, we've been working on it. It's not a leak, it's an error in the metric. We'll finish fixing it, rescore, and put up a thread on the forums. Thanks!</p>",
      "votes": 4,
      "replies": [
        {
          "id": 634006,
          "author_name": "Jack Vial",
          "author_url": "",
          "post_date": "2019-09-25T17:33:17.993000",
          "content": "<p>Great to hear it's not a leak. Looking forward to getting back into this comp. </p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 633950,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-09-25T16:01:34.133000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "633937": "Hi!  Yes, saw this in @joonl04's kernel yesterday, we've been working on it. It's not a leak, it's an error in the metric. We'll finish fixing it, rescore, and put up a thread on the forums. Thanks!",
    "633923": "Guys,\n\nThere is a leak in data or something wrong happened, that's why let's just wait when it is solved. \nJust do not waste of time forking, submitting 0.577 kernels.\n\nAs you have seen, there are 0.577 dummy score with this kernel\nhttps://www.kaggle.com/joonl04/bug-leak-help-us-find-out-0-576-0-577\n\nWhich is obviously just a forked of the original one \nhttps://www.kaggle.com/meaninglesslives/lyft3d-inference-kernel\n\nwith no changes, I think. \n\nBut you can take the original one with 0.231 and changing variables like put something weird ecpochs=25 and get same 0.577. Even higher than in forked version ))) \n\nWhat's the difference between all this 0.577 I don't know, but we played around and managed to get different improvements of \"0.577\"  value)))\n\nSo, let's come back when Lyft will find what's going on )))\n\n ",
    "633950": ""
  }
}