{
  "id": 219091,
  "title": "Regarding Submission Time",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/219091",
  "author_name": "",
  "post_date": "2021-02-13T08:48:52.515520700Z",
  "votes": -3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have submitted my csv file about two hours ago to get the score but it is being run. I can't understand why it is taking too much time to run my submission. Can anyone please help?</p>",
  "messages": [
    {
      "id": "1198733",
      "postDate": "02/13/2021 08:48:52",
      "content": "<p>I have submitted my csv file about two hours ago to get the score but it is being run. I can't understand why it is taking too much time to run my submission. Can anyone please help?</p>",
      "rawMarkdown": "I have submitted my csv file about two hours ago to get the score but it is being run. I can't understand why it is taking too much time to run my submission. Can anyone please help?",
      "votes": null
    },
    {
      "id": "1198741",
      "postDate": "02/13/2021 09:00:00",
      "content": "<p>Hello!</p>\n<p>The test set consists of 15K images, so making predictions actually <strong>should</strong> take a lot of time, especially for large models and ensembles. <br>\nIn your particular case it can be due to many reasons, so please provide a little more information:</p>\n<ul>\n<li>Is your notebook separated into training and inference parts? If not, do so.</li>\n<li>Are you trying to submit huge ensemble predictions? (e.g. submission from 5 EfficientNetB4 ensemble takes about 2 hours on jpegs without TTA)</li>\n<li>Are you using TTA? Each TTA fold means you're adding another 15K images to predict for your model, thus the longer runtime.</li>\n</ul>\n<p>A one-stop solution here is running inference on tfrecords, which takes nearly 2 times faster. </p>",
      "rawMarkdown": "Hello!\n\nThe test set consists of 15K images, so making predictions actually **should** take a lot of time, especially for large models and ensembles. \nIn your particular case it can be due to many reasons, so please provide a little more information:\n* Is your notebook separated into training and inference parts? If not, do so.\n* Are you trying to submit huge ensemble predictions? (e.g. submission from 5 EfficientNetB4 ensemble takes about 2 hours on jpegs without TTA)\n* Are you using TTA? Each TTA fold means you're adding another 15K images to predict for your model, thus the longer runtime.\n\nA one-stop solution here is running inference on tfrecords, which takes nearly 2 times faster.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1198741,
      "author_name": "nickuzmenkov",
      "author_url": "",
      "post_date": "02/13/2021 09:00:00",
      "content": "<p>Hello!</p>\n<p>The test set consists of 15K images, so making predictions actually <strong>should</strong> take a lot of time, especially for large models and ensembles. <br>\nIn your particular case it can be due to many reasons, so please provide a little more information:</p>\n<ul>\n<li>Is your notebook separated into training and inference parts? If not, do so.</li>\n<li>Are you trying to submit huge ensemble predictions? (e.g. submission from 5 EfficientNetB4 ensemble takes about 2 hours on jpegs without TTA)</li>\n<li>Are you using TTA? Each TTA fold means you're adding another 15K images to predict for your model, thus the longer runtime.</li>\n</ul>\n<p>A one-stop solution here is running inference on tfrecords, which takes nearly 2 times faster. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1198733": "I have submitted my csv file about two hours ago to get the score but it is being run. I can't understand why it is taking too much time to run my submission. Can anyone please help?",
    "1198741": "Hello!\n\nThe test set consists of 15K images, so making predictions actually **should** take a lot of time, especially for large models and ensembles. \nIn your particular case it can be due to many reasons, so please provide a little more information:\n* Is your notebook separated into training and inference parts? If not, do so.\n* Are you trying to submit huge ensemble predictions? (e.g. submission from 5 EfficientNetB4 ensemble takes about 2 hours on jpegs without TTA)\n* Are you using TTA? Each TTA fold means you're adding another 15K images to predict for your model, thus the longer runtime.\n\nA one-stop solution here is running inference on tfrecords, which takes nearly 2 times faster."
  },
  "source": "meta"
}