{
  "id": 104909,
  "title": "Submission always timeout",
  "url": "/competitions/aptos2019-blindness-detection/discussion/104909",
  "author_name": "",
  "post_date": "2019-08-20T00:10:00.852340100Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi</p>\n\n<p>Somehow my last 2 submissions both timed out, i don't know how to solve it: I can commit my kernel and generate prediction without problems but when i submit my prediction the kernel always times out.  Does it run my kernel to check the running time against the 9 hour limit? Is it possible to know the details when it times out (it only gives an error message without any detail)?</p>",
  "messages": [
    {
      "id": "603149",
      "postDate": "08/20/2019 00:10:00",
      "content": "<p>Hi</p>\n\n<p>Somehow my last 2 submissions both timed out, i don't know how to solve it: I can commit my kernel and generate prediction without problems but when i submit my prediction the kernel always times out.  Does it run my kernel to check the running time against the 9 hour limit? Is it possible to know the details when it times out (it only gives an error message without any detail)?</p>",
      "rawMarkdown": "Hi\n\nSomehow my last 2 submissions both timed out, i don't know how to solve it: I can commit my kernel and generate prediction without problems but when i submit my prediction the kernel always times out.  Does it run my kernel to check the running time against the 9 hour limit? Is it possible to know the details when it times out (it only gives an error message without any detail)?",
      "votes": null
    },
    {
      "id": "603170",
      "postDate": "08/20/2019 01:12:31",
      "content": "<p>Hi XLCBobby,</p>\n\n<p>See this thread for similar issues:\n<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119</a></p>\n\n<p>I was able to get around this with sort of a hack... I created a new DataFrame for the submission in the sample_submission.csv format.  I made one column titled 'id_code' that has the exact 'id_code' column copied from the test.csv and another column titled 'diagnosis' populated with my integer predictions.  I then used <code>new_df.to_csv(\"submission.csv\", index=False)</code> to save the output file.</p>\n\n<p>Not sure why that worked, but I was finally able to get Kaggle to take the submission.</p>\n\n<p>I hope this helps.  Cheers!</p>",
      "rawMarkdown": "Hi XLCBobby,\n\nSee this thread for similar issues:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119\n\nI was able to get around this with sort of a hack... I created a new DataFrame for the submission in the sample\\_submission.csv format.  I made one column titled 'id\\_code' that has the exact 'id\\_code' column copied from the test.csv and another column titled 'diagnosis' populated with my integer predictions.  I then used `new_df.to_csv(\"submission.csv\", index=False)` to save the output file.\n\nNot sure why that worked, but I was finally able to get Kaggle to take the submission.\n\nI hope this helps.  Cheers!",
      "votes": null
    },
    {
      "id": "603306",
      "postDate": "08/20/2019 05:59:05",
      "content": "<p>Hey Nate thx for reply, i did pretty much the same thing for every submission, didn't help.</p>\n\n<p>Do you know why it would time out? According to the OP of the post you provided above, there's something different with this competition, but i can't find anything other than Kernel run time being 9 hours and no internet connection.</p>",
      "rawMarkdown": "Hey Nate thx for reply, i did pretty much the same thing for every submission, didn't help.\n\nDo you know why it would time out? According to the OP of the post you provided above, there's something different with this competition, but i can't find anything other than Kernel run time being 9 hours and no internet connection.",
      "votes": null
    },
    {
      "id": "603563",
      "postDate": "08/20/2019 12:52:44",
      "content": "<p>The test data you get in your kernel when commit is only 15% of the total test data. And when you submit your kernel, Kaggle runs your kernel for this 15% + the other unreleased 85%. So the submission time cost should be more than (as I experienced) 7 times of your (inference) kernel's time cost. That's probably why your submissions always time out.</p>",
      "rawMarkdown": "The test data you get in your kernel when commit is only 15% of the total test data. And when you submit your kernel, Kaggle runs your kernel for this 15% + the other unreleased 85%. So the submission time cost should be more than (as I experienced) 7 times of your (inference) kernel's time cost. That's probably why your submissions always time out.",
      "votes": null
    },
    {
      "id": "603570",
      "postDate": "08/20/2019 13:01:13",
      "content": "<p>Yeah i think this is more likely to be the reason, i'm using a multi-model ensemble, which makes it much worse.</p>",
      "rawMarkdown": "Yeah i think this is more likely to be the reason, i'm using a multi-model ensemble, which makes it much worse.",
      "votes": null
    },
    {
      "id": "603961",
      "postDate": "08/20/2019 21:52:25",
      "content": "<p>Hey just confirmed yes the multi-model ensemble is the problem, cutting the number of model did the trick. Thanks for your advice!</p>",
      "rawMarkdown": "Hey just confirmed yes the multi-model ensemble is the problem, cutting the number of model did the trick. Thanks for your advice!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 603170,
      "author_name": "cattnb",
      "author_url": "",
      "post_date": "08/20/2019 01:12:31",
      "content": "<p>Hi XLCBobby,</p>\n\n<p>See this thread for similar issues:\n<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119</a></p>\n\n<p>I was able to get around this with sort of a hack... I created a new DataFrame for the submission in the sample_submission.csv format.  I made one column titled 'id_code' that has the exact 'id_code' column copied from the test.csv and another column titled 'diagnosis' populated with my integer predictions.  I then used <code>new_df.to_csv(\"submission.csv\", index=False)</code> to save the output file.</p>\n\n<p>Not sure why that worked, but I was finally able to get Kaggle to take the submission.</p>\n\n<p>I hope this helps.  Cheers!</p>",
      "votes": null,
      "replies": [
        {
          "id": 603306,
          "author_name": "xbldev",
          "author_url": "",
          "post_date": "08/20/2019 05:59:05",
          "content": "<p>Hey Nate thx for reply, i did pretty much the same thing for every submission, didn't help.</p>\n\n<p>Do you know why it would time out? According to the OP of the post you provided above, there's something different with this competition, but i can't find anything other than Kernel run time being 9 hours and no internet connection.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 603563,
      "author_name": "taochen217",
      "author_url": "",
      "post_date": "08/20/2019 12:52:44",
      "content": "<p>The test data you get in your kernel when commit is only 15% of the total test data. And when you submit your kernel, Kaggle runs your kernel for this 15% + the other unreleased 85%. So the submission time cost should be more than (as I experienced) 7 times of your (inference) kernel's time cost. That's probably why your submissions always time out.</p>",
      "votes": null,
      "replies": [
        {
          "id": 603570,
          "author_name": "xbldev",
          "author_url": "",
          "post_date": "08/20/2019 13:01:13",
          "content": "<p>Yeah i think this is more likely to be the reason, i'm using a multi-model ensemble, which makes it much worse.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 603961,
          "author_name": "xbldev",
          "author_url": "",
          "post_date": "08/20/2019 21:52:25",
          "content": "<p>Hey just confirmed yes the multi-model ensemble is the problem, cutting the number of model did the trick. Thanks for your advice!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "603149": "Hi\n\nSomehow my last 2 submissions both timed out, i don't know how to solve it: I can commit my kernel and generate prediction without problems but when i submit my prediction the kernel always times out.  Does it run my kernel to check the running time against the 9 hour limit? Is it possible to know the details when it times out (it only gives an error message without any detail)?",
    "603170": "Hi XLCBobby,\n\nSee this thread for similar issues:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98156#latest-603119\n\nI was able to get around this with sort of a hack... I created a new DataFrame for the submission in the sample\\_submission.csv format.  I made one column titled 'id\\_code' that has the exact 'id\\_code' column copied from the test.csv and another column titled 'diagnosis' populated with my integer predictions.  I then used `new_df.to_csv(\"submission.csv\", index=False)` to save the output file.\n\nNot sure why that worked, but I was finally able to get Kaggle to take the submission.\n\nI hope this helps.  Cheers!",
    "603306": "Hey Nate thx for reply, i did pretty much the same thing for every submission, didn't help.\n\nDo you know why it would time out? According to the OP of the post you provided above, there's something different with this competition, but i can't find anything other than Kernel run time being 9 hours and no internet connection.",
    "603563": "The test data you get in your kernel when commit is only 15% of the total test data. And when you submit your kernel, Kaggle runs your kernel for this 15% + the other unreleased 85%. So the submission time cost should be more than (as I experienced) 7 times of your (inference) kernel's time cost. That's probably why your submissions always time out.",
    "603570": "Yeah i think this is more likely to be the reason, i'm using a multi-model ensemble, which makes it much worse.",
    "603961": "Hey just confirmed yes the multi-model ensemble is the problem, cutting the number of model did the trick. Thanks for your advice!"
  },
  "source": "meta"
}