{
  "id": 75908,
  "title": "Kernel Running time!",
  "url": "/competitions/quora-insincere-questions-classification/discussion/75908",
  "author_name": "",
  "post_date": "2018-12-27T13:07:26.162048600Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hey everyone! I had a couple of very basic questions to ask:\n1) Does everyone test their models on the kaggle kernel itself or some other platform like google colab or personal computers? Is the google colab gpu better than that of kaggle or are both equivalent?\n2) One gets a LB score only after the kernel is committed and the time taken to commit the kernel is very large(in some cases &gt; 1.5 hrs). So does one have to wait that long to check if a tweak in the model has increased/decreased the LB score? Or do we have to make tweaks in the model and test them on the local F1 score itself?</p>\n\n<p>Thank you!</p>",
  "messages": [
    {
      "id": "446074",
      "postDate": "12/27/2018 13:07:26",
      "content": "<p>Hey everyone! I had a couple of very basic questions to ask:\n1) Does everyone test their models on the kaggle kernel itself or some other platform like google colab or personal computers? Is the google colab gpu better than that of kaggle or are both equivalent?\n2) One gets a LB score only after the kernel is committed and the time taken to commit the kernel is very large(in some cases &gt; 1.5 hrs). So does one have to wait that long to check if a tweak in the model has increased/decreased the LB score? Or do we have to make tweaks in the model and test them on the local F1 score itself?</p>\n\n<p>Thank you!</p>",
      "rawMarkdown": "Hey everyone! I had a couple of very basic questions to ask:\n1) Does everyone test their models on the kaggle kernel itself or some other platform like google colab or personal computers? Is the google colab gpu better than that of kaggle or are both equivalent?\n2) One gets a LB score only after the kernel is committed and the time taken to commit the kernel is very large(in some cases &gt; 1.5 hrs). So does one have to wait that long to check if a tweak in the model has increased/decreased the LB score? Or do we have to make tweaks in the model and test them on the local F1 score itself?\n\nThank you!",
      "votes": null
    },
    {
      "id": "446078",
      "postDate": "12/27/2018 13:13:52",
      "content": "<p>Hi Ashish, </p>\n\n<p>You have 2 options:\n1. Create an account for Google Compute or\n2. Save your files after you preprocess them. If you do that you can access those files from the same kernel even after the kernel dies. </p>\n\n<p>Both approaches are sort of useful when you validate using your local CV score. You won't be able to submit to the LB without running the whole kernel in any case, </p>",
      "rawMarkdown": "Hi Ashish, \n\nYou have 2 options:\n1. Create an account for Google Compute or\n2. Save your files after you preprocess them. If you do that you can access those files from the same kernel even after the kernel dies. \n\nBoth approaches are sort of useful when you validate using your local CV score. You won't be able to submit to the LB without running the whole kernel in any case,",
      "votes": null
    },
    {
      "id": "446084",
      "postDate": "12/27/2018 13:23:31",
      "content": "<p><a href=\"/mlwhiz\">@mlwhiz</a> Do you make changes to your model based on the local F1 score or the final LB score(since in case of LB score, you have to wait that long to make any other change)? </p>",
      "rawMarkdown": "mlwhiz Do you make changes to your model based on the local F1 score or the final LB score(since in case of LB score, you have to wait that long to make any other change)?",
      "votes": null
    },
    {
      "id": "446086",
      "postDate": "12/27/2018 13:26:08",
      "content": "<p>In the start, I changed based on LB score but that was the wrong approach. Now I am focussing on my local CV score only. Anyways the LB will shuffle vigorously as soon as stage 2 comes. </p>",
      "rawMarkdown": "In the start, I changed based on LB score but that was the wrong approach. Now I am focussing on my local CV score only. Anyways the LB will shuffle vigorously as soon as stage 2 comes.",
      "votes": null
    },
    {
      "id": "446090",
      "postDate": "12/27/2018 13:32:57",
      "content": "<p><a href=\"/mlwhiz\">@mlwhiz</a> Yes makes sense. Thank you!</p>",
      "rawMarkdown": "mlwhiz Yes makes sense. Thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 446078,
      "author_name": "mlwhiz",
      "author_url": "",
      "post_date": "12/27/2018 13:13:52",
      "content": "<p>Hi Ashish, </p>\n\n<p>You have 2 options:\n1. Create an account for Google Compute or\n2. Save your files after you preprocess them. If you do that you can access those files from the same kernel even after the kernel dies. </p>\n\n<p>Both approaches are sort of useful when you validate using your local CV score. You won't be able to submit to the LB without running the whole kernel in any case, </p>",
      "votes": null,
      "replies": [
        {
          "id": 446084,
          "author_name": "ashish2123",
          "author_url": "",
          "post_date": "12/27/2018 13:23:31",
          "content": "<p><a href=\"/mlwhiz\">@mlwhiz</a> Do you make changes to your model based on the local F1 score or the final LB score(since in case of LB score, you have to wait that long to make any other change)? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 446086,
          "author_name": "mlwhiz",
          "author_url": "",
          "post_date": "12/27/2018 13:26:08",
          "content": "<p>In the start, I changed based on LB score but that was the wrong approach. Now I am focussing on my local CV score only. Anyways the LB will shuffle vigorously as soon as stage 2 comes. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 446090,
          "author_name": "ashish2123",
          "author_url": "",
          "post_date": "12/27/2018 13:32:57",
          "content": "<p><a href=\"/mlwhiz\">@mlwhiz</a> Yes makes sense. Thank you!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "446074": "Hey everyone! I had a couple of very basic questions to ask:\n1) Does everyone test their models on the kaggle kernel itself or some other platform like google colab or personal computers? Is the google colab gpu better than that of kaggle or are both equivalent?\n2) One gets a LB score only after the kernel is committed and the time taken to commit the kernel is very large(in some cases &gt; 1.5 hrs). So does one have to wait that long to check if a tweak in the model has increased/decreased the LB score? Or do we have to make tweaks in the model and test them on the local F1 score itself?\n\nThank you!",
    "446078": "Hi Ashish, \n\nYou have 2 options:\n1. Create an account for Google Compute or\n2. Save your files after you preprocess them. If you do that you can access those files from the same kernel even after the kernel dies. \n\nBoth approaches are sort of useful when you validate using your local CV score. You won't be able to submit to the LB without running the whole kernel in any case,",
    "446084": "mlwhiz Do you make changes to your model based on the local F1 score or the final LB score(since in case of LB score, you have to wait that long to make any other change)?",
    "446086": "In the start, I changed based on LB score but that was the wrong approach. Now I am focussing on my local CV score only. Anyways the LB will shuffle vigorously as soon as stage 2 comes.",
    "446090": "mlwhiz Yes makes sense. Thank you!"
  },
  "source": "meta"
}