{
  "id": 268012,
  "title": "How many full epochs can be run within the time limit ?",
  "url": "/competitions/landmark-recognition-2021/discussion/268012",
  "author_name": "",
  "post_date": "2021-08-25T15:54:38.252350200Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>first epoch for me was showing time around 5 hrs , i used keras image generator class for preprocessing and gpu utilization average was around 15% any suggestions how it could be optimized</p>",
  "messages": [
    {
      "id": "1490371",
      "postDate": "08/25/2021 15:54:38",
      "content": "<p>first epoch for me was showing time around 5 hrs , i used keras image generator class for preprocessing and gpu utilization average was around 15% any suggestions how it could be optimized</p>",
      "rawMarkdown": "first epoch for me was showing time around 5 hrs , i used keras image generator class for preprocessing and gpu utilization average was around 15% any suggestions how it could be optimized",
      "votes": null
    },
    {
      "id": "1491218",
      "postDate": "08/26/2021 08:29:58",
      "content": "<p>This competition have two main problems:</p>\n<ol>\n<li>How to train a classifier able to differentiate between 200,000+ classes within the specified time timit. I suspect that without using TPUs or a GPU cluster it would be impossible.</li>\n<li>How to submit the notebook, which is expected to train the model on both visible and hidden train images, if it throws unspecified errors somewhere within the first minute after submitting. Running and saving the notebook goes smoothly without any errors.</li>\n</ol>\n<p>Try to submit what you have now. It could be a waist of time to train your model any longer. If you manage to submit it without errors, please, explain how to read the data. We will highly appreciate it.</p>",
      "rawMarkdown": "This competition have two main problems:\n1. How to train a classifier able to differentiate between 200,000+ classes within the specified time timit. I suspect that without using TPUs or a GPU cluster it would be impossible.\n2. How to submit the notebook, which is expected to train the model on both visible and hidden train images, if it throws unspecified errors somewhere within the first minute after submitting. Running and saving the notebook goes smoothly without any errors.\n\nTry to submit what you have now. It could be a waist of time to train your model any longer. If you manage to submit it without errors, please, explain how to read the data. We will highly appreciate it.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1491218,
      "author_name": "ekaterinadranitsyna",
      "author_url": "",
      "post_date": "08/26/2021 08:29:58",
      "content": "<p>This competition have two main problems:</p>\n<ol>\n<li>How to train a classifier able to differentiate between 200,000+ classes within the specified time timit. I suspect that without using TPUs or a GPU cluster it would be impossible.</li>\n<li>How to submit the notebook, which is expected to train the model on both visible and hidden train images, if it throws unspecified errors somewhere within the first minute after submitting. Running and saving the notebook goes smoothly without any errors.</li>\n</ol>\n<p>Try to submit what you have now. It could be a waist of time to train your model any longer. If you manage to submit it without errors, please, explain how to read the data. We will highly appreciate it.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1490371": "first epoch for me was showing time around 5 hrs , i used keras image generator class for preprocessing and gpu utilization average was around 15% any suggestions how it could be optimized",
    "1491218": "This competition have two main problems:\n1. How to train a classifier able to differentiate between 200,000+ classes within the specified time timit. I suspect that without using TPUs or a GPU cluster it would be impossible.\n2. How to submit the notebook, which is expected to train the model on both visible and hidden train images, if it throws unspecified errors somewhere within the first minute after submitting. Running and saving the notebook goes smoothly without any errors.\n\nTry to submit what you have now. It could be a waist of time to train your model any longer. If you manage to submit it without errors, please, explain how to read the data. We will highly appreciate it."
  },
  "source": "meta"
}