{
  "id": 103420,
  "title": "Uploading your own pre-trained models",
  "url": "/competitions/aptos2019-blindness-detection/discussion/103420",
  "author_name": "",
  "post_date": "2019-08-09T06:11:17.462068400Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi all, </p>\n\n<p>First time Kaggle kernels user here.</p>\n\n<p>I was reading this in the competition description:</p>\n\n<p><em>You can still train a model offline, upload it as a dataset, and use the kernel exclusively to perform inference</em></p>\n\n<p>Just want to clarify, does this mean I can train my model offline, upload the model .pth file and then use that to generate predictions rather than doing the training on the Kernel directly? </p>\n\n<p>I thought the instruction to \"upload dataset\" was confusing so just wanted to clarify that my thinking was on track and that I was able to upload an actual model file?</p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "595358",
      "postDate": "08/09/2019 06:11:17",
      "content": "<p>Hi all, </p>\n\n<p>First time Kaggle kernels user here.</p>\n\n<p>I was reading this in the competition description:</p>\n\n<p><em>You can still train a model offline, upload it as a dataset, and use the kernel exclusively to perform inference</em></p>\n\n<p>Just want to clarify, does this mean I can train my model offline, upload the model .pth file and then use that to generate predictions rather than doing the training on the Kernel directly? </p>\n\n<p>I thought the instruction to \"upload dataset\" was confusing so just wanted to clarify that my thinking was on track and that I was able to upload an actual model file?</p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "Hi all, \n\nFirst time Kaggle kernels user here.\n\nI was reading this in the competition description:\n\n*You can still train a model offline, upload it as a dataset, and use the kernel exclusively to perform inference*\n\nJust want to clarify, does this mean I can train my model offline, upload the model .pth file and then use that to generate predictions rather than doing the training on the Kernel directly? \n\nI thought the instruction to \"upload dataset\" was confusing so just wanted to clarify that my thinking was on track and that I was able to upload an actual model file?\n\nThanks!",
      "votes": null
    },
    {
      "id": "595386",
      "postDate": "08/09/2019 07:00:21",
      "content": "<p>Absolutely right!</p>",
      "rawMarkdown": "Absolutely right!",
      "votes": null
    },
    {
      "id": "595463",
      "postDate": "08/09/2019 08:58:44",
      "content": "<p>Yes, that's correct understanding. You upload your model as Dataset, add it to your kernel workspace , load the model to predict :)</p>",
      "rawMarkdown": "Yes, that's correct understanding. You upload your model as Dataset, add it to your kernel workspace , load the model to predict :)",
      "votes": null
    },
    {
      "id": "596178",
      "postDate": "08/10/2019 08:59:14",
      "content": "<p>Thank you everyone! Got it working now :) </p>",
      "rawMarkdown": "Thank you everyone! Got it working now :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 595386,
      "author_name": "oceanwong",
      "author_url": "",
      "post_date": "08/09/2019 07:00:21",
      "content": "<p>Absolutely right!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 595463,
      "author_name": "harshthaker",
      "author_url": "",
      "post_date": "08/09/2019 08:58:44",
      "content": "<p>Yes, that's correct understanding. You upload your model as Dataset, add it to your kernel workspace , load the model to predict :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 596178,
      "author_name": "adeperio",
      "author_url": "",
      "post_date": "08/10/2019 08:59:14",
      "content": "<p>Thank you everyone! Got it working now :) </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "595358": "Hi all, \n\nFirst time Kaggle kernels user here.\n\nI was reading this in the competition description:\n\n*You can still train a model offline, upload it as a dataset, and use the kernel exclusively to perform inference*\n\nJust want to clarify, does this mean I can train my model offline, upload the model .pth file and then use that to generate predictions rather than doing the training on the Kernel directly? \n\nI thought the instruction to \"upload dataset\" was confusing so just wanted to clarify that my thinking was on track and that I was able to upload an actual model file?\n\nThanks!",
    "595386": "Absolutely right!",
    "595463": "Yes, that's correct understanding. You upload your model as Dataset, add it to your kernel workspace , load the model to predict :)",
    "596178": "Thank you everyone! Got it working now :)"
  },
  "source": "meta"
}