{
  "id": 135780,
  "title": "I wish that the error submission does not count as 1 of the 5 daily submissions",
  "url": "/competitions/bengaliai-cv19/discussion/135780",
  "author_name": "",
  "post_date": "2020-03-15T23:58:32.579181900Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Given that:\n- the code ran properly with the sample test data. \n- the code is also tested on the full training data set to make sure the kernel runs correctly on large data sets \n- There is no error trace other than \"Notebook Threw Exception\"/</p>\n\n<p>There could be many reasons that cause the issue, after multiple trials I think it's likely due to some edge input case that breaks the pre-processing pipeline. Unfortunately, 5 tries a day is not enough for me to narrow down which part cause the issue. , and there are 10 tries left. </p>\n\n<p>I do understand you want to make sure people don't spam submissions, I wish there could be some error trace or some flexibility on submission numbers for error cases.</p>",
  "messages": [
    {
      "id": "773816",
      "postDate": "03/15/2020 23:58:32",
      "content": "<p>Given that:\n- the code ran properly with the sample test data. \n- the code is also tested on the full training data set to make sure the kernel runs correctly on large data sets \n- There is no error trace other than \"Notebook Threw Exception\"/</p>\n\n<p>There could be many reasons that cause the issue, after multiple trials I think it's likely due to some edge input case that breaks the pre-processing pipeline. Unfortunately, 5 tries a day is not enough for me to narrow down which part cause the issue. , and there are 10 tries left. </p>\n\n<p>I do understand you want to make sure people don't spam submissions, I wish there could be some error trace or some flexibility on submission numbers for error cases.</p>",
      "rawMarkdown": "Given that:\n- the code ran properly with the sample test data. \n- the code is also tested on the full training data set to make sure the kernel runs correctly on large data sets \n- There is no error trace other than \"Notebook Threw Exception\"/\n\nThere could be many reasons that cause the issue, after multiple trials I think it's likely due to some edge input case that breaks the pre-processing pipeline. Unfortunately, 5 tries a day is not enough for me to narrow down which part cause the issue. , and there are 10 tries left. \n\nI do understand you want to make sure people don't spam submissions, I wish there could be some error trace or some flexibility on submission numbers for error cases.",
      "votes": null
    },
    {
      "id": "775510",
      "postDate": "03/16/2020 18:49:35",
      "content": "<p>If you believe its only an edge case, wrap your preprocessing in try except blocks, atleast you can get the score nearby to your actual score with the few images missed. It would also be a good idea to run your preprocessing code on the entire train set to see if something breaks.</p>",
      "rawMarkdown": "If you believe its only an edge case, wrap your preprocessing in try except blocks, atleast you can get the score nearby to your actual score with the few images missed. It would also be a good idea to run your preprocessing code on the entire train set to see if something breaks.",
      "votes": null
    },
    {
      "id": "775690",
      "postDate": "03/17/2020 00:05:30",
      "content": "<p>that's exactly what i did, output [0,0,0] in case of error, and I finally made one successful submission.</p>\n\n<p>my model on 10 folds cross validation/ 5 folds validation gives kaggle score of 97.6, 97.5.... but ended up with 96.89 on public, 93.7 on private... I suspect there maybe multiple edge cases that broke my code... or my model really sucks T_T..</p>",
      "rawMarkdown": "that's exactly what i did, output [0,0,0] in case of error, and I finally made one successful submission.\n\nmy model on 10 folds cross validation/ 5 folds validation gives kaggle score of 97.6, 97.5.... but ended up with 96.89 on public, 93.7 on private... I suspect there maybe multiple edge cases that broke my code... or my model really sucks T_T..",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 775510,
      "author_name": "dipamc77",
      "author_url": "",
      "post_date": "03/16/2020 18:49:35",
      "content": "<p>If you believe its only an edge case, wrap your preprocessing in try except blocks, atleast you can get the score nearby to your actual score with the few images missed. It would also be a good idea to run your preprocessing code on the entire train set to see if something breaks.</p>",
      "votes": null,
      "replies": [
        {
          "id": 775690,
          "author_name": "mingyangzhou",
          "author_url": "",
          "post_date": "03/17/2020 00:05:30",
          "content": "<p>that's exactly what i did, output [0,0,0] in case of error, and I finally made one successful submission.</p>\n\n<p>my model on 10 folds cross validation/ 5 folds validation gives kaggle score of 97.6, 97.5.... but ended up with 96.89 on public, 93.7 on private... I suspect there maybe multiple edge cases that broke my code... or my model really sucks T_T..</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "773816": "Given that:\n- the code ran properly with the sample test data. \n- the code is also tested on the full training data set to make sure the kernel runs correctly on large data sets \n- There is no error trace other than \"Notebook Threw Exception\"/\n\nThere could be many reasons that cause the issue, after multiple trials I think it's likely due to some edge input case that breaks the pre-processing pipeline. Unfortunately, 5 tries a day is not enough for me to narrow down which part cause the issue. , and there are 10 tries left. \n\nI do understand you want to make sure people don't spam submissions, I wish there could be some error trace or some flexibility on submission numbers for error cases.",
    "775510": "If you believe its only an edge case, wrap your preprocessing in try except blocks, atleast you can get the score nearby to your actual score with the few images missed. It would also be a good idea to run your preprocessing code on the entire train set to see if something breaks.",
    "775690": "that's exactly what i did, output [0,0,0] in case of error, and I finally made one successful submission.\n\nmy model on 10 folds cross validation/ 5 folds validation gives kaggle score of 97.6, 97.5.... but ended up with 96.89 on public, 93.7 on private... I suspect there maybe multiple edge cases that broke my code... or my model really sucks T_T.."
  },
  "source": "meta"
}