{
  "id": 107267,
  "title": "I'm concerning huge shake-up",
  "url": "/competitions/aptos2019-blindness-detection/discussion/107267",
  "author_name": "",
  "post_date": "2019-09-03T10:42:06.746417300Z",
  "votes": 16,
  "comment_count": 14,
  "views": 0,
  "content": "<p>As you know, train and test are somewhat different.</p>\n\n<p>So, I guess there might be huge shake-up. </p>\n\n<p>Because we have 9 hours to inference and submit, I suggest you all prepare the submissions in safe way to avoid shake-up.  (can stack many models)</p>\n\n<p>And be careful to make a notebook(kernel) which run in 9 hours. \nSafely, 8hour 30 minute kernel is good to use!</p>\n\n<p>I think that how to safely ensemble the models is key to win this competition.</p>\n\n<p>Good luck your models :)</p>",
  "messages": [
    {
      "id": "616659",
      "postDate": "09/03/2019 10:42:06",
      "content": "<p>As you know, train and test are somewhat different.</p>\n\n<p>So, I guess there might be huge shake-up. </p>\n\n<p>Because we have 9 hours to inference and submit, I suggest you all prepare the submissions in safe way to avoid shake-up.  (can stack many models)</p>\n\n<p>And be careful to make a notebook(kernel) which run in 9 hours. \nSafely, 8hour 30 minute kernel is good to use!</p>\n\n<p>I think that how to safely ensemble the models is key to win this competition.</p>\n\n<p>Good luck your models :)</p>",
      "rawMarkdown": "As you know, train and test are somewhat different.\n\nSo, I guess there might be huge shake-up. \n\nBecause we have 9 hours to inference and submit, I suggest you all prepare the submissions in safe way to avoid shake-up.  (can stack many models)\n\nAnd be careful to make a notebook(kernel) which run in 9 hours. \nSafely, 8hour 30 minute kernel is good to use!\n\nI think that how to safely ensemble the models is key to win this competition.\n\nGood luck your models :)",
      "votes": null
    },
    {
      "id": "617494",
      "postDate": "09/04/2019 07:25:12",
      "content": "<p>Did you mean shake-up in running time ? Yesterday, my 20 minutes kernels ran for 3 hours. If this happens in the final run, then all running time calculations could go wrong.</p>",
      "rawMarkdown": "Did you mean shake-up in running time ? Yesterday, my 20 minutes kernels ran for 3 hours. If this happens in the final run, then all running time calculations could go wrong.",
      "votes": null
    },
    {
      "id": "617711",
      "postDate": "09/04/2019 12:03:14",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F682501%2F96406c156bf71c93449494bda6a39163%2F1?generation=1567598585302612&amp;alt=media\" alt=\"\"></p>\n\n<p>Haven't taken this competition, but according to the previous one, the leaderboard is stable. Wish you all goog luck !!!</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F682501%2F96406c156bf71c93449494bda6a39163%2F1?generation=1567598585302612&amp;alt=media)\n\nHaven't taken this competition, but according to the previous one, the leaderboard is stable. Wish you all goog luck !!!",
      "votes": null
    },
    {
      "id": "617903",
      "postDate": "09/04/2019 15:40:19",
      "content": "<p>I see, the distributions are different for training and pb set, and the number of samples are much less than before. This competition is hard : (</p>",
      "rawMarkdown": "I see, the distributions are different for training and pb set, and the number of samples are much less than before. This competition is hard : (",
      "votes": null
    },
    {
      "id": "617908",
      "postDate": "09/04/2019 15:48:26",
      "content": "<p>True ! </p>",
      "rawMarkdown": "True !",
      "votes": null
    },
    {
      "id": "617983",
      "postDate": "09/04/2019 17:26:10",
      "content": "<p>believe in yourself.. whosover know what they are doing and why are they doing  are generally fundamentally correct  so will not see huge difference.. in contrary any one getting scores by hook n crook will find its right place :)</p>",
      "rawMarkdown": "believe in yourself.. whosover know what they are doing and why are they doing  are generally fundamentally correct  so will not see huge difference.. in contrary any one getting scores by hook n crook will find its right place :)",
      "votes": null
    },
    {
      "id": "618403",
      "postDate": "09/05/2019 07:21:51",
      "content": "<p>I mean the shake-up in leaderboard.  The number of images of public test and private test is different, so that your running time is getting larger!</p>",
      "rawMarkdown": "I mean the shake-up in leaderboard.  The number of images of public test and private test is different, so that your running time is getting larger!",
      "votes": null
    },
    {
      "id": "618427",
      "postDate": "09/05/2019 07:44:54",
      "content": "<p>Got it. Private dataset is much larger so there could be shake-ups. Let's see. I initially thought you wanted to point out this issue that people are facing: <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129</a></p>",
      "rawMarkdown": "Got it. Private dataset is much larger so there could be shake-ups. Let's see. I initially thought you wanted to point out this issue that people are facing: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129",
      "votes": null
    },
    {
      "id": "618445",
      "postDate": "09/05/2019 08:06:54",
      "content": "<p>I think if our submissions have succeeded once, we need not care about private run time.\nBecause this competition is synchronized one.</p>",
      "rawMarkdown": "I think if our submissions have succeeded once, we need not care about private run time.\nBecause this competition is synchronized one.",
      "votes": null
    },
    {
      "id": "618452",
      "postDate": "09/05/2019 08:14:21",
      "content": "<p>agree, if we submit successfully and get a score,  which shows that our kernel works well on private test.</p>",
      "rawMarkdown": "agree, if we submit successfully and get a score,  which shows that our kernel works well on private test.",
      "votes": null
    },
    {
      "id": "618786",
      "postDate": "09/05/2019 13:45:56",
      "content": "<p>Hello <a href=\"/garybios\">@garybios</a>, do you mean that the submit button for the preliminary score already uses the private set of ~13000 samples?</p>\n\n<p>Thank you,\nJesús</p>",
      "rawMarkdown": "Hello @garybios, do you mean that the submit button for the preliminary score already uses the private set of ~13000 samples?\n\nThank you,\nJesús",
      "votes": null
    },
    {
      "id": "618813",
      "postDate": "09/05/2019 14:23:29",
      "content": "<p>yes, it does</p>",
      "rawMarkdown": "yes, it does",
      "votes": null
    },
    {
      "id": "618820",
      "postDate": "09/05/2019 14:34:59",
      "content": "<p>Hello <a href=\"/kvlmll\">@kvlmll</a>,</p>\n\n<p>How do you infer this? It's not what I understand from the instructions in <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/data\">https://www.kaggle.com/c/aptos2019-blindness-detection/data</a></p>\n\n<p>I thought the public score is based on the public test set only. My pre-processing algorithm is very time consuming and I won't be able to get a private score, but I was hopeful for the public one.</p>\n\n<p>Thank you,\nJesús</p>",
      "rawMarkdown": "Hello @kvlmll,\n\nHow do you infer this? It's not what I understand from the instructions in https://www.kaggle.com/c/aptos2019-blindness-detection/data\n\nI thought the public score is based on the public test set only. My pre-processing algorithm is very time consuming and I won't be able to get a private score, but I was hopeful for the public one.\n\nThank you,\nJesús",
      "votes": null
    },
    {
      "id": "618829",
      "postDate": "09/05/2019 14:42:56",
      "content": "<p>I think this has been explained quite a few times now: You click commit, the kernel uses only the public part, you click submit, the kernel uses the full part (public+private). Only if it runs through properly in submit, you can expect to get a score in private LB.</p>",
      "rawMarkdown": "I think this has been explained quite a few times now: You click commit, the kernel uses only the public part, you click submit, the kernel uses the full part (public+private). Only if it runs through properly in submit, you can expect to get a score in private LB.",
      "votes": null
    },
    {
      "id": "618834",
      "postDate": "09/05/2019 14:49:43",
      "content": "<p>I'm sorry, I did't find in my searches. Now is perfectly explained. Thank you <a href=\"/philippsinger\">@philippsinger</a> </p>",
      "rawMarkdown": "I'm sorry, I did't find in my searches. Now is perfectly explained. Thank you @philippsinger",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 617494,
      "author_name": "harshthaker",
      "author_url": "",
      "post_date": "09/04/2019 07:25:12",
      "content": "<p>Did you mean shake-up in running time ? Yesterday, my 20 minutes kernels ran for 3 hours. If this happens in the final run, then all running time calculations could go wrong.</p>",
      "votes": null,
      "replies": [
        {
          "id": 618403,
          "author_name": "youhanlee",
          "author_url": "",
          "post_date": "09/05/2019 07:21:51",
          "content": "<p>I mean the shake-up in leaderboard.  The number of images of public test and private test is different, so that your running time is getting larger!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618427,
          "author_name": "harshthaker",
          "author_url": "",
          "post_date": "09/05/2019 07:44:54",
          "content": "<p>Got it. Private dataset is much larger so there could be shake-ups. Let's see. I initially thought you wanted to point out this issue that people are facing: <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 617711,
      "author_name": "yiheng",
      "author_url": "",
      "post_date": "09/04/2019 12:03:14",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F682501%2F96406c156bf71c93449494bda6a39163%2F1?generation=1567598585302612&amp;alt=media\" alt=\"\"></p>\n\n<p>Haven't taken this competition, but according to the previous one, the leaderboard is stable. Wish you all goog luck !!!</p>",
      "votes": null,
      "replies": [
        {
          "id": 617903,
          "author_name": "yiheng",
          "author_url": "",
          "post_date": "09/04/2019 15:40:19",
          "content": "<p>I see, the distributions are different for training and pb set, and the number of samples are much less than before. This competition is hard : (</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 617908,
          "author_name": "harshthaker",
          "author_url": "",
          "post_date": "09/04/2019 15:48:26",
          "content": "<p>True ! </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 617983,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "09/04/2019 17:26:10",
          "content": "<p>believe in yourself.. whosover know what they are doing and why are they doing  are generally fundamentally correct  so will not see huge difference.. in contrary any one getting scores by hook n crook will find its right place :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 618445,
      "author_name": "maxwell110",
      "author_url": "",
      "post_date": "09/05/2019 08:06:54",
      "content": "<p>I think if our submissions have succeeded once, we need not care about private run time.\nBecause this competition is synchronized one.</p>",
      "votes": null,
      "replies": [
        {
          "id": 618452,
          "author_name": "garybios",
          "author_url": "",
          "post_date": "09/05/2019 08:14:21",
          "content": "<p>agree, if we submit successfully and get a score,  which shows that our kernel works well on private test.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618786,
          "author_name": "jmartindelasierra",
          "author_url": "",
          "post_date": "09/05/2019 13:45:56",
          "content": "<p>Hello <a href=\"/garybios\">@garybios</a>, do you mean that the submit button for the preliminary score already uses the private set of ~13000 samples?</p>\n\n<p>Thank you,\nJesús</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618813,
          "author_name": "kvlmll",
          "author_url": "",
          "post_date": "09/05/2019 14:23:29",
          "content": "<p>yes, it does</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618820,
          "author_name": "jmartindelasierra",
          "author_url": "",
          "post_date": "09/05/2019 14:34:59",
          "content": "<p>Hello <a href=\"/kvlmll\">@kvlmll</a>,</p>\n\n<p>How do you infer this? It's not what I understand from the instructions in <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/data\">https://www.kaggle.com/c/aptos2019-blindness-detection/data</a></p>\n\n<p>I thought the public score is based on the public test set only. My pre-processing algorithm is very time consuming and I won't be able to get a private score, but I was hopeful for the public one.</p>\n\n<p>Thank you,\nJesús</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618829,
          "author_name": "philippsinger",
          "author_url": "",
          "post_date": "09/05/2019 14:42:56",
          "content": "<p>I think this has been explained quite a few times now: You click commit, the kernel uses only the public part, you click submit, the kernel uses the full part (public+private). Only if it runs through properly in submit, you can expect to get a score in private LB.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618834,
          "author_name": "jmartindelasierra",
          "author_url": "",
          "post_date": "09/05/2019 14:49:43",
          "content": "<p>I'm sorry, I did't find in my searches. Now is perfectly explained. Thank you <a href=\"/philippsinger\">@philippsinger</a> </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "616659": "As you know, train and test are somewhat different.\n\nSo, I guess there might be huge shake-up. \n\nBecause we have 9 hours to inference and submit, I suggest you all prepare the submissions in safe way to avoid shake-up.  (can stack many models)\n\nAnd be careful to make a notebook(kernel) which run in 9 hours. \nSafely, 8hour 30 minute kernel is good to use!\n\nI think that how to safely ensemble the models is key to win this competition.\n\nGood luck your models :)",
    "617494": "Did you mean shake-up in running time ? Yesterday, my 20 minutes kernels ran for 3 hours. If this happens in the final run, then all running time calculations could go wrong.",
    "617711": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F682501%2F96406c156bf71c93449494bda6a39163%2F1?generation=1567598585302612&amp;alt=media)\n\nHaven't taken this competition, but according to the previous one, the leaderboard is stable. Wish you all goog luck !!!",
    "617903": "I see, the distributions are different for training and pb set, and the number of samples are much less than before. This competition is hard : (",
    "617908": "True !",
    "617983": "believe in yourself.. whosover know what they are doing and why are they doing  are generally fundamentally correct  so will not see huge difference.. in contrary any one getting scores by hook n crook will find its right place :)",
    "618403": "I mean the shake-up in leaderboard.  The number of images of public test and private test is different, so that your running time is getting larger!",
    "618427": "Got it. Private dataset is much larger so there could be shake-ups. Let's see. I initially thought you wanted to point out this issue that people are facing: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-618129",
    "618445": "I think if our submissions have succeeded once, we need not care about private run time.\nBecause this competition is synchronized one.",
    "618452": "agree, if we submit successfully and get a score,  which shows that our kernel works well on private test.",
    "618786": "Hello @garybios, do you mean that the submit button for the preliminary score already uses the private set of ~13000 samples?\n\nThank you,\nJesús",
    "618813": "yes, it does",
    "618820": "Hello @kvlmll,\n\nHow do you infer this? It's not what I understand from the instructions in https://www.kaggle.com/c/aptos2019-blindness-detection/data\n\nI thought the public score is based on the public test set only. My pre-processing algorithm is very time consuming and I won't be able to get a private score, but I was hopeful for the public one.\n\nThank you,\nJesús",
    "618829": "I think this has been explained quite a few times now: You click commit, the kernel uses only the public part, you click submit, the kernel uses the full part (public+private). Only if it runs through properly in submit, you can expect to get a score in private LB.",
    "618834": "I'm sorry, I did't find in my searches. Now is perfectly explained. Thank you @philippsinger"
  },
  "source": "meta"
}