{
  "id": 374538,
  "title": "Question about kaggle notebooks for beginer",
  "url": "/competitions/otto-recommender-system/discussion/374538",
  "author_name": "",
  "post_date": "2022-12-27T19:20:47.449929500Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi. Guys. </p>\n<p>Although I joined kaggle 6 years ago, I'm rather new to kaggle competition and I have limited experimence in kaggle competition.</p>\n<p>I have always had a question about the kaggle notebooks, and hope your guys can give me some advices:  <strong>is kaggle notebooks enough for model iteration? especially when data is big, and when you want to try several different method?</strong></p>\n<p>The reason that I have this question is that, in my work or when I was working on a kaggle competition for practice, I always use Pycharm, I will create debug_data to quickly validate my code, and I will also create Python Class for train, predict and eval, in addition, the pipeline should be configurable easily to test different types of model.  All these are for quick validation and quick iterations. </p>\n<p>With all the codes in a notebook, and with such a big data, I think it makes iteration so slow. How do you guys cope with these problems? </p>\n<p>Again, I have limited experience on kaggle, maybe my ideas are wrong. I'm happy to hear about your  opinions </p>\n<p>Thanks, guys.</p>",
  "messages": [
    {
      "id": "2077662",
      "postDate": "12/27/2022 19:20:47",
      "content": "<p>Hi. Guys. </p>\n<p>Although I joined kaggle 6 years ago, I'm rather new to kaggle competition and I have limited experimence in kaggle competition.</p>\n<p>I have always had a question about the kaggle notebooks, and hope your guys can give me some advices:  <strong>is kaggle notebooks enough for model iteration? especially when data is big, and when you want to try several different method?</strong></p>\n<p>The reason that I have this question is that, in my work or when I was working on a kaggle competition for practice, I always use Pycharm, I will create debug_data to quickly validate my code, and I will also create Python Class for train, predict and eval, in addition, the pipeline should be configurable easily to test different types of model.  All these are for quick validation and quick iterations. </p>\n<p>With all the codes in a notebook, and with such a big data, I think it makes iteration so slow. How do you guys cope with these problems? </p>\n<p>Again, I have limited experience on kaggle, maybe my ideas are wrong. I'm happy to hear about your  opinions </p>\n<p>Thanks, guys.</p>",
      "rawMarkdown": "Hi. Guys. \n\nAlthough I joined kaggle 6 years ago, I'm rather new to kaggle competition and I have limited experimence in kaggle competition.\n\nI have always had a question about the kaggle notebooks, and hope your guys can give me some advices:  **is kaggle notebooks enough for model iteration? especially when data is big, and when you want to try several different method?**\n\n\nThe reason that I have this question is that, in my work or when I was working on a kaggle competition for practice, I always use Pycharm, I will create debug_data to quickly validate my code, and I will also create Python Class for train, predict and eval, in addition, the pipeline should be configurable easily to test different types of model.  All these are for quick validation and quick iterations. \n\nWith all the codes in a notebook, and with such a big data, I think it makes iteration so slow. How do you guys cope with these problems? \n\n\nAgain, I have limited experience on kaggle, maybe my ideas are wrong. I'm happy to hear about your  opinions \n\nThanks, guys.",
      "votes": null
    },
    {
      "id": "2077665",
      "postDate": "12/27/2022 19:24:22",
      "content": "<p>Usually I use Kaggle notebooks for small simple EDA and models that do not necessitate other GPU/ TPU usage. Please note that Kaggle offers very limited GPU quotas that get exhausted very quickly. You may perhaps expose the dataset to another cloud platform or your local device and then train a model suitably, using the GPUs at your disposal. </p>\n<p>Hope this helps <a href=\"https://www.kaggle.com/huaguo\" target=\"_blank\">@huaguo</a> </p>",
      "rawMarkdown": "Usually I use Kaggle notebooks for small simple EDA and models that do not necessitate other GPU/ TPU usage. Please note that Kaggle offers very limited GPU quotas that get exhausted very quickly. You may perhaps expose the dataset to another cloud platform or your local device and then train a model suitably, using the GPUs at your disposal. \n\nHope this helps @huaguo",
      "votes": null
    },
    {
      "id": "2077675",
      "postDate": "12/27/2022 19:37:39",
      "content": "<p>Hi.  Ravi. Thanks.  </p>\n<p>One more question, when you works with your local machine or cloud, do you use scripts more or notebooks?   Do you have a template project structure to create your project with Pycharm or VScode?</p>",
      "rawMarkdown": "Hi.  Ravi. Thanks.  \n\nOne more question, when you works with your local machine or cloud, do you use scripts more or notebooks?   Do you have a template project structure to create your project with Pycharm or VScode?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2077665,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "12/27/2022 19:24:22",
      "content": "<p>Usually I use Kaggle notebooks for small simple EDA and models that do not necessitate other GPU/ TPU usage. Please note that Kaggle offers very limited GPU quotas that get exhausted very quickly. You may perhaps expose the dataset to another cloud platform or your local device and then train a model suitably, using the GPUs at your disposal. </p>\n<p>Hope this helps <a href=\"https://www.kaggle.com/huaguo\" target=\"_blank\">@huaguo</a> </p>",
      "votes": null,
      "replies": [
        {
          "id": 2077675,
          "author_name": "huaguo",
          "author_url": "",
          "post_date": "12/27/2022 19:37:39",
          "content": "<p>Hi.  Ravi. Thanks.  </p>\n<p>One more question, when you works with your local machine or cloud, do you use scripts more or notebooks?   Do you have a template project structure to create your project with Pycharm or VScode?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2077662": "Hi. Guys. \n\nAlthough I joined kaggle 6 years ago, I'm rather new to kaggle competition and I have limited experimence in kaggle competition.\n\nI have always had a question about the kaggle notebooks, and hope your guys can give me some advices:  **is kaggle notebooks enough for model iteration? especially when data is big, and when you want to try several different method?**\n\n\nThe reason that I have this question is that, in my work or when I was working on a kaggle competition for practice, I always use Pycharm, I will create debug_data to quickly validate my code, and I will also create Python Class for train, predict and eval, in addition, the pipeline should be configurable easily to test different types of model.  All these are for quick validation and quick iterations. \n\nWith all the codes in a notebook, and with such a big data, I think it makes iteration so slow. How do you guys cope with these problems? \n\n\nAgain, I have limited experience on kaggle, maybe my ideas are wrong. I'm happy to hear about your  opinions \n\nThanks, guys.",
    "2077665": "Usually I use Kaggle notebooks for small simple EDA and models that do not necessitate other GPU/ TPU usage. Please note that Kaggle offers very limited GPU quotas that get exhausted very quickly. You may perhaps expose the dataset to another cloud platform or your local device and then train a model suitably, using the GPUs at your disposal. \n\nHope this helps @huaguo",
    "2077675": "Hi.  Ravi. Thanks.  \n\nOne more question, when you works with your local machine or cloud, do you use scripts more or notebooks?   Do you have a template project structure to create your project with Pycharm or VScode?"
  },
  "source": "meta"
}