{
  "id": 116461,
  "title": "Can I upload my offline prediction file as dataset, then read it in kernel and make submission?",
  "url": "/competitions/tensorflow2-question-answering/discussion/116461",
  "author_name": "Zan Shuxun",
  "post_date": "2019-11-09T09:08:52.094000",
  "votes": 2,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I don't understand the rule \"all your submissions will be made from a Kaggle Kernel\". Now that we can train the model offline and upload the trained model, why not make submission offline?</p>\n\n<p>Making predictions in kernel requires many additional packages, there are too many scripts needs to add.</p>",
  "messages": [
    {
      "id": 669145,
      "postDate": "2019-11-09T15:05:40.617Z",
      "content": "<p>You can train your models offline and upload it as a dataset and make predictions to the test set in kaggle kernels. Let's say you are using tf2.0 framework to do the training then after training you will get an <code>.h5</code> model file which you can upload and <code>load</code> it in kaggle kernels to make predictions on test set</p>",
      "rawMarkdown": "You can train your models offline and upload it as a dataset and make predictions to the test set in kaggle kernels. Let's say you are using tf2.0 framework to do the training then after training you will get an `.h5` model file which you can upload and `load` it in kaggle kernels to make predictions on test set",
      "votes": 3,
      "replies": [
        {
          "id": 669505,
          "postDate": "2019-11-10T05:47:33.980Z",
          "content": "<p>Thank you. But I used many additional packages to run the models. Can I make predictions offline and uplaod the submission.csv as a dataset? It's very troublesome to add the scripts into kaggle kernel.</p>",
          "rawMarkdown": "Thank you. But I used many additional packages to run the models. Can I make predictions offline and uplaod the submission.csv as a dataset? It's very troublesome to add the scripts into kaggle kernel."
        },
        {
          "id": 669649,
          "postDate": "2019-11-10T09:39:30.133Z",
          "content": "<blockquote>\n  <p>Can I make predictions offline and uplaod the submission.csv as a dataset?</p>\n</blockquote>\n\n<p>I am not sure if you can do that as this seems to be Kernels Only Competition, which means when you use your model to make predictions, you are making predictions both for public test and private test set. If you make predictions offline and upload the submission.csv then you are only making predictions for the public test, not private test set</p>",
          "rawMarkdown": "&gt; Can I make predictions offline and uplaod the submission.csv as a dataset?\n\nI am not sure if you can do that as this seems to be Kernels Only Competition, which means when you use your model to make predictions, you are making predictions both for public test and private test set. If you make predictions offline and upload the submission.csv then you are only making predictions for the public test, not private test set\n",
          "votes": 2
        },
        {
          "id": 669835,
          "postDate": "2019-11-10T14:46:40.827Z",
          "content": "<p>Bibek is correct. If you write code that only writes predictions for the public test set, it will fail when rerun on the private test set, which is completely withheld. The point of a code competition is that you write your model in Kaggle’s notebooks environment so that we can rerun that code on the private test set without handing it out.</p>",
          "rawMarkdown": "Bibek is correct. If you write code that only writes predictions for the public test set, it will fail when rerun on the private test set, which is completely withheld. The point of a code competition is that you write your model in Kaggle’s notebooks environment so that we can rerun that code on the private test set without handing it out.",
          "votes": 1
        },
        {
          "id": 671714,
          "postDate": "2019-11-13T04:33:23.147Z",
          "content": "<p>Get it. Thanks! </p>",
          "rawMarkdown": "Get it. Thanks! "
        },
        {
          "id": 671715,
          "postDate": "2019-11-13T04:33:48.613Z",
          "content": "<p>Thanks a lot!</p>",
          "rawMarkdown": "Thanks a lot!"
        }
      ]
    },
    {
      "id": 668996,
      "postDate": "2019-11-09T09:08:52.093Z",
      "content": "<p>I don't understand the rule \"all your submissions will be made from a Kaggle Kernel\". Now that we can train the model offline and upload the trained model, why not make submission offline?</p>\n\n<p>Making predictions in kernel requires many additional packages, there are too many scripts needs to add.</p>",
      "rawMarkdown": "I don't understand the rule \"all your submissions will be made from a Kaggle Kernel\". Now that we can train the model offline and upload the trained model, why not make submission offline?\n\nMaking predictions in kernel requires many additional packages, there are too many scripts needs to add.",
      "votes": 2
    },
    {
      "id": 668999,
      "postDate": "2019-11-09T09:12:58.930Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 669145,
      "author_name": "Bibek",
      "author_url": "",
      "post_date": "2019-11-09T15:05:40.617000",
      "content": "<p>You can train your models offline and upload it as a dataset and make predictions to the test set in kaggle kernels. Let's say you are using tf2.0 framework to do the training then after training you will get an <code>.h5</code> model file which you can upload and <code>load</code> it in kaggle kernels to make predictions on test set</p>",
      "votes": 3,
      "replies": [
        {
          "id": 669505,
          "author_name": "Zan Shuxun",
          "author_url": "",
          "post_date": "2019-11-10T05:47:33.980000",
          "content": "<p>Thank you. But I used many additional packages to run the models. Can I make predictions offline and uplaod the submission.csv as a dataset? It's very troublesome to add the scripts into kaggle kernel.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 669649,
          "author_name": "Bibek",
          "author_url": "",
          "post_date": "2019-11-10T09:39:30.133000",
          "content": "<blockquote>\n  <p>Can I make predictions offline and uplaod the submission.csv as a dataset?</p>\n</blockquote>\n\n<p>I am not sure if you can do that as this seems to be Kernels Only Competition, which means when you use your model to make predictions, you are making predictions both for public test and private test set. If you make predictions offline and upload the submission.csv then you are only making predictions for the public test, not private test set</p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 669835,
          "author_name": "Julia Elliott",
          "author_url": "",
          "post_date": "2019-11-10T14:46:40.827000",
          "content": "<p>Bibek is correct. If you write code that only writes predictions for the public test set, it will fail when rerun on the private test set, which is completely withheld. The point of a code competition is that you write your model in Kaggle’s notebooks environment so that we can rerun that code on the private test set without handing it out.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 671714,
          "author_name": "Zan Shuxun",
          "author_url": "",
          "post_date": "2019-11-13T04:33:23.147000",
          "content": "<p>Get it. Thanks! </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 671715,
          "author_name": "Zan Shuxun",
          "author_url": "",
          "post_date": "2019-11-13T04:33:48.613000",
          "content": "<p>Thanks a lot!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 668999,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-11-09T09:12:58.930000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "669145": "You can train your models offline and upload it as a dataset and make predictions to the test set in kaggle kernels. Let's say you are using tf2.0 framework to do the training then after training you will get an `.h5` model file which you can upload and `load` it in kaggle kernels to make predictions on test set",
    "668996": "I don't understand the rule \"all your submissions will be made from a Kaggle Kernel\". Now that we can train the model offline and upload the trained model, why not make submission offline?\n\nMaking predictions in kernel requires many additional packages, there are too many scripts needs to add.",
    "668999": ""
  }
}