{
  "id": 270560,
  "title": "Will pesudo-probing submissions also cause disqualification ?",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/270560",
  "author_name": "",
  "post_date": "2021-09-06T04:57:09.007298200Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Making 5 submissions from notebook with restricted GPU time is hard, so i have been submitting the result csv obtained from local environment in a hard code format. </p>\n<p>This is because i dont want to wait for notebook execution in firstly saving version for hours and then waiting for submission rerun for hours. before getting results. Submitting hardcode gives result in seconds.</p>\n<p>i was hoping to reproduce results on kaggle notebook once i am satisfied, but apparently others have used hardcode format for probing.  </p>\n<p>i read the rules, it says nothing about hard coded submissions, just probing, i am concerned if my technique would lie under probing umbrella. </p>\n<p>Insights Anyone ?</p>",
  "messages": [
    {
      "id": "1504101",
      "postDate": "09/06/2021 04:57:09",
      "content": "<p>Making 5 submissions from notebook with restricted GPU time is hard, so i have been submitting the result csv obtained from local environment in a hard code format. </p>\n<p>This is because i dont want to wait for notebook execution in firstly saving version for hours and then waiting for submission rerun for hours. before getting results. Submitting hardcode gives result in seconds.</p>\n<p>i was hoping to reproduce results on kaggle notebook once i am satisfied, but apparently others have used hardcode format for probing.  </p>\n<p>i read the rules, it says nothing about hard coded submissions, just probing, i am concerned if my technique would lie under probing umbrella. </p>\n<p>Insights Anyone ?</p>",
      "rawMarkdown": "Making 5 submissions from notebook with restricted GPU time is hard, so i have been submitting the result csv obtained from local environment in a hard code format. \n\nThis is because i dont want to wait for notebook execution in firstly saving version for hours and then waiting for submission rerun for hours. before getting results. Submitting hardcode gives result in seconds.\n\ni was hoping to reproduce results on kaggle notebook once i am satisfied, but apparently others have used hardcode format for probing.  \n\ni read the rules, it says nothing about hard coded submissions, just probing, i am concerned if my technique would lie under probing umbrella. \n\nInsights Anyone ?",
      "votes": null
    },
    {
      "id": "1504636",
      "postDate": "09/06/2021 14:27:13",
      "content": "<p>This technique shouldn't be against the rules, although i'm not sure. How i combat this issue is by having separate training and prediction notebooks. While training takes roughly 6-7 hours, prediction notebooks complete within minutes.</p>",
      "rawMarkdown": "This technique shouldn't be against the rules, although i'm not sure. How i combat this issue is by having separate training and prediction notebooks. While training takes roughly 6-7 hours, prediction notebooks complete within minutes.",
      "votes": null
    },
    {
      "id": "1504881",
      "postDate": "09/06/2021 18:25:26",
      "content": "<p>I am trying same technique, unfortunately preprocessing test data on my end takes 8.5 hrs on submission, preds take seconds<br>\nWith 87 samples it takes 1.3 hrs while saving new version only </p>",
      "rawMarkdown": "I am trying same technique, unfortunately preprocessing test data on my end takes 8.5 hrs on submission, preds take seconds\nWith 87 samples it takes 1.3 hrs while saving new version only",
      "votes": null
    },
    {
      "id": "1505552",
      "postDate": "09/07/2021 11:12:22",
      "content": "<p>This is so far the best technique i know of. By saving model files, train once and test multiple times, without time constraints. This also helps in ensembling models by trying out different permutations and combinations from your trained models.</p>",
      "rawMarkdown": "This is so far the best technique i know of. By saving model files, train once and test multiple times, without time constraints. This also helps in ensembling models by trying out different permutations and combinations from your trained models.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1504636,
      "author_name": "snish9",
      "author_url": "",
      "post_date": "09/06/2021 14:27:13",
      "content": "<p>This technique shouldn't be against the rules, although i'm not sure. How i combat this issue is by having separate training and prediction notebooks. While training takes roughly 6-7 hours, prediction notebooks complete within minutes.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1504881,
          "author_name": "shummi39",
          "author_url": "",
          "post_date": "09/06/2021 18:25:26",
          "content": "<p>I am trying same technique, unfortunately preprocessing test data on my end takes 8.5 hrs on submission, preds take seconds<br>\nWith 87 samples it takes 1.3 hrs while saving new version only </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1505552,
          "author_name": "snish9",
          "author_url": "",
          "post_date": "09/07/2021 11:12:22",
          "content": "<p>This is so far the best technique i know of. By saving model files, train once and test multiple times, without time constraints. This also helps in ensembling models by trying out different permutations and combinations from your trained models.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1504101": "Making 5 submissions from notebook with restricted GPU time is hard, so i have been submitting the result csv obtained from local environment in a hard code format. \n\nThis is because i dont want to wait for notebook execution in firstly saving version for hours and then waiting for submission rerun for hours. before getting results. Submitting hardcode gives result in seconds.\n\ni was hoping to reproduce results on kaggle notebook once i am satisfied, but apparently others have used hardcode format for probing.  \n\ni read the rules, it says nothing about hard coded submissions, just probing, i am concerned if my technique would lie under probing umbrella. \n\nInsights Anyone ?",
    "1504636": "This technique shouldn't be against the rules, although i'm not sure. How i combat this issue is by having separate training and prediction notebooks. While training takes roughly 6-7 hours, prediction notebooks complete within minutes.",
    "1504881": "I am trying same technique, unfortunately preprocessing test data on my end takes 8.5 hrs on submission, preds take seconds\nWith 87 samples it takes 1.3 hrs while saving new version only",
    "1505552": "This is so far the best technique i know of. By saving model files, train once and test multiple times, without time constraints. This also helps in ensembling models by trying out different permutations and combinations from your trained models."
  },
  "source": "meta"
}