{
  "id": 107430,
  "title": "Kernel runs for 1h but after commit returns Timeout",
  "url": "/competitions/aptos2019-blindness-detection/discussion/107430",
  "author_name": "",
  "post_date": "2019-09-04T10:21:44.110797600Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Good day everybody!</p>\n\n<p>I have seen the posts on forum about strange kernels behaviour recently. So I want to find out whether my situation is connected to it, or is it normal. I have a kernel and want to submit it. Inside it, I measure a time of run - and I see that it completes in ~1 hour. However, when I submit it to competition, after ~9 hours it returns Kernel Timeout. So does it mean that the private data is so big? Seems strange. As I see from the public leaderboard:</p>\n\n<p>&gt; This leaderboard is calculated with approximately 15% of the test data. The final results will be based on the other 85%, so the final standings may be different.</p>\n\n<p>So assuming that the time spent on each sample is relatively similar, we have that kernel runs on 15% of the test data for ~60 min, so in total it should be maximum ~60min / 0.15 = 400 min &lt; 7 hours &lt; required 9 hours. Why does it output kernel timeout then?</p>",
  "messages": [
    {
      "id": "617623",
      "postDate": "09/04/2019 10:21:44",
      "content": "<p>Good day everybody!</p>\n\n<p>I have seen the posts on forum about strange kernels behaviour recently. So I want to find out whether my situation is connected to it, or is it normal. I have a kernel and want to submit it. Inside it, I measure a time of run - and I see that it completes in ~1 hour. However, when I submit it to competition, after ~9 hours it returns Kernel Timeout. So does it mean that the private data is so big? Seems strange. As I see from the public leaderboard:</p>\n\n<p>&gt; This leaderboard is calculated with approximately 15% of the test data. The final results will be based on the other 85%, so the final standings may be different.</p>\n\n<p>So assuming that the time spent on each sample is relatively similar, we have that kernel runs on 15% of the test data for ~60 min, so in total it should be maximum ~60min / 0.15 = 400 min &lt; 7 hours &lt; required 9 hours. Why does it output kernel timeout then?</p>",
      "rawMarkdown": "Good day everybody!\n\nI have seen the posts on forum about strange kernels behaviour recently. So I want to find out whether my situation is connected to it, or is it normal. I have a kernel and want to submit it. Inside it, I measure a time of run - and I see that it completes in ~1 hour. However, when I submit it to competition, after ~9 hours it returns Kernel Timeout. So does it mean that the private data is so big? Seems strange. As I see from the public leaderboard:\n\n&gt; This leaderboard is calculated with approximately 15% of the test data. The final results will be based on the other 85%, so the final standings may be different.\n\nSo assuming that the time spent on each sample is relatively similar, we have that kernel runs on 15% of the test data for ~60 min, so in total it should be maximum ~60min / 0.15 = 400 min &lt; 7 hours &lt; required 9 hours. Why does it output kernel timeout then?",
      "votes": null
    },
    {
      "id": "617645",
      "postDate": "09/04/2019 10:46:05",
      "content": "<p>look at this thread <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447</a>\n- If you submitted today morning, probably you should try to submit it one more time</p>\n\n<p>If step 1 fails then:</p>\n\n<ul>\n<li>Try to reduce the number of your ensemble models (if you did so), and test it. If it succeeds, then the problem is in here.</li>\n</ul>",
      "rawMarkdown": "look at this thread https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447\n- If you submitted today morning, probably you should try to submit it one more time\n\nIf step 1 fails then:\n\n- Try to reduce the number of your ensemble models (if you did so), and test it. If it succeeds, then the problem is in here.",
      "votes": null
    },
    {
      "id": "618115",
      "postDate": "09/04/2019 21:11:24",
      "content": "<p>Thank you, you were right - seems like the problem is in the number of models. However, I still don't understand why it runs for so long, taking into the account that on the visible part of data the kernel runs for one hour.</p>",
      "rawMarkdown": "Thank you, you were right - seems like the problem is in the number of models. However, I still don't understand why it runs for so long, taking into the account that on the visible part of data the kernel runs for one hour.",
      "votes": null
    },
    {
      "id": "618171",
      "postDate": "09/05/2019 00:36:21",
      "content": "<p>Hope it helps. </p>\n\n<p>BTW: the problem started again or kaggle team did not fix it completely, cause one of my submissions did not finish and Timeout Error appeared, although I could submit the same submission before. </p>",
      "rawMarkdown": "Hope it helps. \n\nBTW: the problem started again or kaggle team did not fix it completely, cause one of my submissions did not finish and Timeout Error appeared, although I could submit the same submission before.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 617645,
      "author_name": "nuller",
      "author_url": "",
      "post_date": "09/04/2019 10:46:05",
      "content": "<p>look at this thread <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447</a>\n- If you submitted today morning, probably you should try to submit it one more time</p>\n\n<p>If step 1 fails then:</p>\n\n<ul>\n<li>Try to reduce the number of your ensemble models (if you did so), and test it. If it succeeds, then the problem is in here.</li>\n</ul>",
      "votes": null,
      "replies": [
        {
          "id": 618115,
          "author_name": "blackitten13",
          "author_url": "",
          "post_date": "09/04/2019 21:11:24",
          "content": "<p>Thank you, you were right - seems like the problem is in the number of models. However, I still don't understand why it runs for so long, taking into the account that on the visible part of data the kernel runs for one hour.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 618171,
          "author_name": "nuller",
          "author_url": "",
          "post_date": "09/05/2019 00:36:21",
          "content": "<p>Hope it helps. </p>\n\n<p>BTW: the problem started again or kaggle team did not fix it completely, cause one of my submissions did not finish and Timeout Error appeared, although I could submit the same submission before. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "617623": "Good day everybody!\n\nI have seen the posts on forum about strange kernels behaviour recently. So I want to find out whether my situation is connected to it, or is it normal. I have a kernel and want to submit it. Inside it, I measure a time of run - and I see that it completes in ~1 hour. However, when I submit it to competition, after ~9 hours it returns Kernel Timeout. So does it mean that the private data is so big? Seems strange. As I see from the public leaderboard:\n\n&gt; This leaderboard is calculated with approximately 15% of the test data. The final results will be based on the other 85%, so the final standings may be different.\n\nSo assuming that the time spent on each sample is relatively similar, we have that kernel runs on 15% of the test data for ~60 min, so in total it should be maximum ~60min / 0.15 = 400 min &lt; 7 hours &lt; required 9 hours. Why does it output kernel timeout then?",
    "617645": "look at this thread https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107323#latest-617447\n- If you submitted today morning, probably you should try to submit it one more time\n\nIf step 1 fails then:\n\n- Try to reduce the number of your ensemble models (if you did so), and test it. If it succeeds, then the problem is in here.",
    "618115": "Thank you, you were right - seems like the problem is in the number of models. However, I still don't understand why it runs for so long, taking into the account that on the visible part of data the kernel runs for one hour.",
    "618171": "Hope it helps. \n\nBTW: the problem started again or kaggle team did not fix it completely, cause one of my submissions did not finish and Timeout Error appeared, although I could submit the same submission before."
  },
  "source": "meta"
}