{
  "id": 174317,
  "title": "How to make final submission using ensembles",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/174317",
  "author_name": "",
  "post_date": "2020-08-13T06:03:21.332667800Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have been ensembling many models to get a better score. For final submission, is it fine to just load all the submission files and submit it? I dont think it will work since these submission files does not contain the 70% held out test set.<br>\nShould we load models and make predictions on test dataset in our final submission notebook</p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": "968589",
      "postDate": "08/13/2020 06:03:21",
      "content": "<p>I have been ensembling many models to get a better score. For final submission, is it fine to just load all the submission files and submit it? I dont think it will work since these submission files does not contain the 70% held out test set.<br>\nShould we load models and make predictions on test dataset in our final submission notebook</p>\n<p>Thanks</p>",
      "rawMarkdown": "I have been ensembling many models to get a better score. For final submission, is it fine to just load all the submission files and submit it? I dont think it will work since these submission files does not contain the 70% held out test set.\nShould we load models and make predictions on test dataset in our final submission notebook\n\nThanks",
      "votes": null
    },
    {
      "id": "968732",
      "postDate": "08/13/2020 07:56:06",
      "content": "<p>ensembling submissions is fine. Kaggle calculates a secret score for all of your submissions, which is used for the private lb result, so they don't have to recalculate in the end and each submission includes predictions on the whole test set.</p>",
      "rawMarkdown": "ensembling submissions is fine. Kaggle calculates a secret score for all of your submissions, which is used for the private lb result, so they don't have to recalculate in the end and each submission includes predictions on the whole test set.",
      "votes": null
    },
    {
      "id": "970087",
      "postDate": "08/14/2020 07:27:05",
      "content": "<p>Thanks Roman. <br>\nSo, basically I can create datasets for all my submissions and load them in the final notebook.</p>",
      "rawMarkdown": "Thanks Roman. \nSo, basically I can create datasets for all my submissions and load them in the final notebook.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 968732,
      "author_name": "romanweilguny",
      "author_url": "",
      "post_date": "08/13/2020 07:56:06",
      "content": "<p>ensembling submissions is fine. Kaggle calculates a secret score for all of your submissions, which is used for the private lb result, so they don't have to recalculate in the end and each submission includes predictions on the whole test set.</p>",
      "votes": null,
      "replies": [
        {
          "id": 970087,
          "author_name": "ajaykumar7778",
          "author_url": "",
          "post_date": "08/14/2020 07:27:05",
          "content": "<p>Thanks Roman. <br>\nSo, basically I can create datasets for all my submissions and load them in the final notebook.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "968589": "I have been ensembling many models to get a better score. For final submission, is it fine to just load all the submission files and submit it? I dont think it will work since these submission files does not contain the 70% held out test set.\nShould we load models and make predictions on test dataset in our final submission notebook\n\nThanks",
    "968732": "ensembling submissions is fine. Kaggle calculates a secret score for all of your submissions, which is used for the private lb result, so they don't have to recalculate in the end and each submission includes predictions on the whole test set.",
    "970087": "Thanks Roman. \nSo, basically I can create datasets for all my submissions and load them in the final notebook."
  },
  "source": "meta"
}