{
  "id": 177759,
  "title": "Need help understanding the competition (the test set and the purpose of images)",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/177759",
  "author_name": "",
  "post_date": "2020-08-27T09:17:18.478200100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Here's my understanding -</p>\n<p>We are supposed to predict the decline in <code>FVC</code> starting from <code>week 0</code> (rather <code>week -12</code>) until <code>week 133</code>. We have the <code>initial FVC</code>, i.e. <code>FVC</code> at <code>week 0</code>.</p>\n<ol>\n<li><p>The patients in test already exist in train. Which means we have information about their FVC decline. I tried regressing just those patients (using the meta data) and predicting <code>FVC</code>, but that didn't work. Not surprising, it couldn't be that straight forward. Please help me understand what's wrong with this approach?</p></li>\n<li><p>For the images, we don't know which image is for which week (we don't, right?). How do the images help in the prediction based on <code>Week 0</code>? I reckon we can use <code>week 0</code> image for predicting decline, what about the rest of images? I guess, we could use each image as an individual instance and see the rate of decline thereon. and use that for predicting based on <code>week 0</code> image.</p></li>\n</ol>\n<p>Most likely, these are naive questions and I am sure many have already answered them in the posts and notebooks, somehow I still couldn't wrap my head around it. </p>\n<p>Any help is much appreciated. Thanks..!!</p>",
  "messages": [
    {
      "id": "987448",
      "postDate": "08/27/2020 09:17:18",
      "content": "<p>Here's my understanding -</p>\n<p>We are supposed to predict the decline in <code>FVC</code> starting from <code>week 0</code> (rather <code>week -12</code>) until <code>week 133</code>. We have the <code>initial FVC</code>, i.e. <code>FVC</code> at <code>week 0</code>.</p>\n<ol>\n<li><p>The patients in test already exist in train. Which means we have information about their FVC decline. I tried regressing just those patients (using the meta data) and predicting <code>FVC</code>, but that didn't work. Not surprising, it couldn't be that straight forward. Please help me understand what's wrong with this approach?</p></li>\n<li><p>For the images, we don't know which image is for which week (we don't, right?). How do the images help in the prediction based on <code>Week 0</code>? I reckon we can use <code>week 0</code> image for predicting decline, what about the rest of images? I guess, we could use each image as an individual instance and see the rate of decline thereon. and use that for predicting based on <code>week 0</code> image.</p></li>\n</ol>\n<p>Most likely, these are naive questions and I am sure many have already answered them in the posts and notebooks, somehow I still couldn't wrap my head around it. </p>\n<p>Any help is much appreciated. Thanks..!!</p>",
      "rawMarkdown": "Here's my understanding -\n\nWe are supposed to predict the decline in `FVC` starting from `week 0` (rather `week -12`) until `week 133`. We have the `initial FVC`, i.e. `FVC` at `week 0`.\n\n1. The patients in test already exist in train. Which means we have information about their FVC decline. I tried regressing just those patients (using the meta data) and predicting `FVC`, but that didn't work. Not surprising, it couldn't be that straight forward. Please help me understand what's wrong with this approach?\n\n2. For the images, we don't know which image is for which week (we don't, right?). How do the images help in the prediction based on `Week 0`? I reckon we can use `week 0` image for predicting decline, what about the rest of images? I guess, we could use each image as an individual instance and see the rate of decline thereon. and use that for predicting based on `week 0` image.\n\nMost likely, these are naive questions and I am sure many have already answered them in the posts and notebooks, somehow I still couldn't wrap my head around it. \n\nAny help is much appreciated. Thanks..!!",
      "votes": null
    },
    {
      "id": "987487",
      "postDate": "08/27/2020 09:54:50",
      "content": "<p><a href=\"https://www.kaggle.com/oshanm\" target=\"_blank\">@oshanm</a> </p>\n<ol>\n<li><p>First of all the test patients provided for us is just for you to check if the submission notebook runs properly and if the predictions are being saved in the correct format. You get a leaderboard score when your notebook is rerun with a hidden test set when you submit your submission.csv file.</p></li>\n<li><p>The image set for each patient is a single CT scan image consisting of multiple slices taken at week 0. The different files in each folder with Patient_id as the folder name is just 1 big image. You can take a look at the following discussions to get an idea of how to use the images with tabular data.</p></li>\n</ol>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/165727</a><br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/166123</a></p>",
      "rawMarkdown": "oshanm \n1. First of all the test patients provided for us is just for you to check if the submission notebook runs properly and if the predictions are being saved in the correct format. You get a leaderboard score when your notebook is rerun with a hidden test set when you submit your submission.csv file.\n\n2. The image set for each patient is a single CT scan image consisting of multiple slices taken at week 0. The different files in each folder with Patient_id as the folder name is just 1 big image. You can take a look at the following discussions to get an idea of how to use the images with tabular data.\n\n[https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/165727](url)\n[https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/166123](url)",
      "votes": null
    },
    {
      "id": "987525",
      "postDate": "08/27/2020 10:22:29",
      "content": "<p>Thanks <a href=\"https://www.kaggle.com/yovinyahathugoda\" target=\"_blank\">@yovinyahathugoda</a> .  This is helpful. </p>",
      "rawMarkdown": "Thanks @yovinyahathugoda .  This is helpful.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 987487,
      "author_name": "yovinyahathugoda",
      "author_url": "",
      "post_date": "08/27/2020 09:54:50",
      "content": "<p><a href=\"https://www.kaggle.com/oshanm\" target=\"_blank\">@oshanm</a> </p>\n<ol>\n<li><p>First of all the test patients provided for us is just for you to check if the submission notebook runs properly and if the predictions are being saved in the correct format. You get a leaderboard score when your notebook is rerun with a hidden test set when you submit your submission.csv file.</p></li>\n<li><p>The image set for each patient is a single CT scan image consisting of multiple slices taken at week 0. The different files in each folder with Patient_id as the folder name is just 1 big image. You can take a look at the following discussions to get an idea of how to use the images with tabular data.</p></li>\n</ol>\n<p><a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/165727</a><br>\n<a href=\"url\" target=\"_blank\">https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/166123</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 987525,
          "author_name": "oshanm",
          "author_url": "",
          "post_date": "08/27/2020 10:22:29",
          "content": "<p>Thanks <a href=\"https://www.kaggle.com/yovinyahathugoda\" target=\"_blank\">@yovinyahathugoda</a> .  This is helpful. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "987448": "Here's my understanding -\n\nWe are supposed to predict the decline in `FVC` starting from `week 0` (rather `week -12`) until `week 133`. We have the `initial FVC`, i.e. `FVC` at `week 0`.\n\n1. The patients in test already exist in train. Which means we have information about their FVC decline. I tried regressing just those patients (using the meta data) and predicting `FVC`, but that didn't work. Not surprising, it couldn't be that straight forward. Please help me understand what's wrong with this approach?\n\n2. For the images, we don't know which image is for which week (we don't, right?). How do the images help in the prediction based on `Week 0`? I reckon we can use `week 0` image for predicting decline, what about the rest of images? I guess, we could use each image as an individual instance and see the rate of decline thereon. and use that for predicting based on `week 0` image.\n\nMost likely, these are naive questions and I am sure many have already answered them in the posts and notebooks, somehow I still couldn't wrap my head around it. \n\nAny help is much appreciated. Thanks..!!",
    "987487": "oshanm \n1. First of all the test patients provided for us is just for you to check if the submission notebook runs properly and if the predictions are being saved in the correct format. You get a leaderboard score when your notebook is rerun with a hidden test set when you submit your submission.csv file.\n\n2. The image set for each patient is a single CT scan image consisting of multiple slices taken at week 0. The different files in each folder with Patient_id as the folder name is just 1 big image. You can take a look at the following discussions to get an idea of how to use the images with tabular data.\n\n[https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/165727](url)\n[https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/166123](url)",
    "987525": "Thanks @yovinyahathugoda .  This is helpful."
  },
  "source": "meta"
}