{
  "id": 101089,
  "title": "Reducing Committing Time of kernel",
  "url": "/competitions/aptos2019-blindness-detection/discussion/101089",
  "author_name": "Ashutosh Hathidara",
  "post_date": "2019-07-23T06:54:44.873000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have observed that many of the participants are facing problems because the kernels in this competition is taking very much time to commit. I have also faced the same issue.</p>\n\n<p><strong>Solution :</strong>\nI have observed in some of kernels which belongs to some of the grandmasters. They are preprocessing and training the data on the local machine and building the model file. They are adding the model file on the remote kernel and writing the code only for test data for preprocessing and for generating predictions.</p>\n\n<p>This will reduce the commit time significantly since at the time of commit, whole notebook is running again. \nI have tried with my solution. With train + test code, it was taking <strong>6 hours</strong> to commit. But I have tried above approach. Now it is taking approx. <strong>2.5 hours</strong> .</p>\n\n<p>Thank you !\nAll the Best !</p>",
  "messages": [
    {
      "id": 582545,
      "postDate": "2019-07-23T10:04:30.897Z",
      "content": "<p>Commit Time can even be reduced to few seconds if you produce an error using size of test data and still kernel will run without error when submitting. </p>",
      "rawMarkdown": "Commit Time can even be reduced to few seconds if you produce an error using size of test data and still kernel will run without error when submitting. "
    },
    {
      "id": 582421,
      "postDate": "2019-07-23T06:54:44.873Z",
      "content": "<p>I have observed that many of the participants are facing problems because the kernels in this competition is taking very much time to commit. I have also faced the same issue.</p>\n\n<p><strong>Solution :</strong>\nI have observed in some of kernels which belongs to some of the grandmasters. They are preprocessing and training the data on the local machine and building the model file. They are adding the model file on the remote kernel and writing the code only for test data for preprocessing and for generating predictions.</p>\n\n<p>This will reduce the commit time significantly since at the time of commit, whole notebook is running again. \nI have tried with my solution. With train + test code, it was taking <strong>6 hours</strong> to commit. But I have tried above approach. Now it is taking approx. <strong>2.5 hours</strong> .</p>\n\n<p>Thank you !\nAll the Best !</p>",
      "rawMarkdown": "I have observed that many of the participants are facing problems because the kernels in this competition is taking very much time to commit. I have also faced the same issue.\n\n**Solution :**\nI have observed in some of kernels which belongs to some of the grandmasters. They are preprocessing and training the data on the local machine and building the model file. They are adding the model file on the remote kernel and writing the code only for test data for preprocessing and for generating predictions.\n\nThis will reduce the commit time significantly since at the time of commit, whole notebook is running again. \nI have tried with my solution. With train + test code, it was taking **6 hours** to commit. But I have tried above approach. Now it is taking approx. **2.5 hours** .\n\nThank you !\nAll the Best !"
    },
    {
      "id": 582837,
      "postDate": "2019-07-23T16:47:26.370Z",
      "rawMarkdown": "",
      "votes": -6,
      "isDeleted": true
    },
    {
      "id": 582836,
      "postDate": "2019-07-23T16:47:03.617Z",
      "rawMarkdown": "",
      "votes": -5,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 582545,
      "author_name": "Cold",
      "author_url": "",
      "post_date": "2019-07-23T10:04:30.897000",
      "content": "<p>Commit Time can even be reduced to few seconds if you produce an error using size of test data and still kernel will run without error when submitting. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 582837,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-07-23T16:47:26.370000",
      "content": "",
      "votes": -6,
      "replies": []
    },
    {
      "id": 582836,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-07-23T16:47:03.617000",
      "content": "",
      "votes": -5,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "582545": "Commit Time can even be reduced to few seconds if you produce an error using size of test data and still kernel will run without error when submitting. ",
    "582421": "I have observed that many of the participants are facing problems because the kernels in this competition is taking very much time to commit. I have also faced the same issue.\n\n**Solution :**\nI have observed in some of kernels which belongs to some of the grandmasters. They are preprocessing and training the data on the local machine and building the model file. They are adding the model file on the remote kernel and writing the code only for test data for preprocessing and for generating predictions.\n\nThis will reduce the commit time significantly since at the time of commit, whole notebook is running again. \nI have tried with my solution. With train + test code, it was taking **6 hours** to commit. But I have tried above approach. Now it is taking approx. **2.5 hours** .\n\nThank you !\nAll the Best !",
    "582837": "",
    "582836": ""
  }
}