{
  "id": 160791,
  "title": "Efficiently using cost-efficient cloud GPU's/TPU's for Kaggle competitions",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/160791",
  "author_name": "Rohit Agarwal",
  "post_date": "2020-06-22T17:18:29.490000",
  "votes": 0,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I'm a noob here and I'm trying to figure out a reliable cost-efficient way of using cloud to compete in Kaggle competitions. Earlier, the competition organisers used to provide GCP credits to users, however that doesn't seem to be the case nowadays. I am not able to use my local machine for model training since I don't have NVIDIA GPU's. I have currently taken up Google Colab Pro subscription, however it doesn't offer persistent storage. Sure, I can store files in Google Drive and mount it in colab, however the I/O is very slow and breaks easily. Also, the kernel stops if the browser is inactive for a while. I would have loved to use Kaggle Kernels, however the strict time limitations means model training can't happen for large number of epochs.</p>\n\n<p>What would you recommend?</p>",
  "messages": [
    {
      "id": 897184,
      "postDate": "2020-06-22T17:18:29.490Z",
      "content": "<p>I'm a noob here and I'm trying to figure out a reliable cost-efficient way of using cloud to compete in Kaggle competitions. Earlier, the competition organisers used to provide GCP credits to users, however that doesn't seem to be the case nowadays. I am not able to use my local machine for model training since I don't have NVIDIA GPU's. I have currently taken up Google Colab Pro subscription, however it doesn't offer persistent storage. Sure, I can store files in Google Drive and mount it in colab, however the I/O is very slow and breaks easily. Also, the kernel stops if the browser is inactive for a while. I would have loved to use Kaggle Kernels, however the strict time limitations means model training can't happen for large number of epochs.</p>\n\n<p>What would you recommend?</p>",
      "rawMarkdown": "I'm a noob here and I'm trying to figure out a reliable cost-efficient way of using cloud to compete in Kaggle competitions. Earlier, the competition organisers used to provide GCP credits to users, however that doesn't seem to be the case nowadays. I am not able to use my local machine for model training since I don't have NVIDIA GPU's. I have currently taken up Google Colab Pro subscription, however it doesn't offer persistent storage. Sure, I can store files in Google Drive and mount it in colab, however the I/O is very slow and breaks easily. Also, the kernel stops if the browser is inactive for a while. I would have loved to use Kaggle Kernels, however the strict time limitations means model training can't happen for large number of epochs.\n\nWhat would you recommend?"
    },
    {
      "id": 897970,
      "postDate": "2020-06-23T07:48:57.927Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 897970,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-06-23T07:48:57.927000",
      "content": "",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "897184": "I'm a noob here and I'm trying to figure out a reliable cost-efficient way of using cloud to compete in Kaggle competitions. Earlier, the competition organisers used to provide GCP credits to users, however that doesn't seem to be the case nowadays. I am not able to use my local machine for model training since I don't have NVIDIA GPU's. I have currently taken up Google Colab Pro subscription, however it doesn't offer persistent storage. Sure, I can store files in Google Drive and mount it in colab, however the I/O is very slow and breaks easily. Also, the kernel stops if the browser is inactive for a while. I would have loved to use Kaggle Kernels, however the strict time limitations means model training can't happen for large number of epochs.\n\nWhat would you recommend?",
    "897970": ""
  }
}