{
  "id": 164771,
  "title": "Can I use several kernels in my pipeline and just submit one?",
  "url": "/competitions/birdsong-recognition/discussion/164771",
  "author_name": "",
  "post_date": "2020-07-07T13:34:28.346515200Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I thought that a code competition (having to submit kernels) would impose a limit for the time taken by my pipeline (from preprocessing to inference). But now I see that one can use several kernels for preprocessing, then use these kernels outputs as inputs for another kernel for training, so on and so forth, and just submit the final inference kernel. </p>\n\n<p>Is this a valid approach? Can I use as many kernels as I want and just submit the last one?</p>",
  "messages": [
    {
      "id": "918769",
      "postDate": "07/07/2020 13:34:28",
      "content": "<p>I thought that a code competition (having to submit kernels) would impose a limit for the time taken by my pipeline (from preprocessing to inference). But now I see that one can use several kernels for preprocessing, then use these kernels outputs as inputs for another kernel for training, so on and so forth, and just submit the final inference kernel. </p>\n\n<p>Is this a valid approach? Can I use as many kernels as I want and just submit the last one?</p>",
      "rawMarkdown": "I thought that a code competition (having to submit kernels) would impose a limit for the time taken by my pipeline (from preprocessing to inference). But now I see that one can use several kernels for preprocessing, then use these kernels outputs as inputs for another kernel for training, so on and so forth, and just submit the final inference kernel. \n\nIs this a valid approach? Can I use as many kernels as I want and just submit the last one?",
      "votes": null
    },
    {
      "id": "918876",
      "postDate": "07/07/2020 14:57:14",
      "content": "<p>You can use as many kernels for training and experimentation. Your final submitted kernel needs to have the inference pipeline.</p>\n\n<p>Note that preprocessing the test data (and any other processing of test data) will also need to be part of the final kernel because only the submitted kernel can access the entire test data.</p>",
      "rawMarkdown": "You can use as many kernels for training and experimentation. Your final submitted kernel needs to have the inference pipeline.\n\nNote that preprocessing the test data (and any other processing of test data) will also need to be part of the final kernel because only the submitted kernel can access the entire test data.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 918876,
      "author_name": "rohanrao",
      "author_url": "",
      "post_date": "07/07/2020 14:57:14",
      "content": "<p>You can use as many kernels for training and experimentation. Your final submitted kernel needs to have the inference pipeline.</p>\n\n<p>Note that preprocessing the test data (and any other processing of test data) will also need to be part of the final kernel because only the submitted kernel can access the entire test data.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "918769": "I thought that a code competition (having to submit kernels) would impose a limit for the time taken by my pipeline (from preprocessing to inference). But now I see that one can use several kernels for preprocessing, then use these kernels outputs as inputs for another kernel for training, so on and so forth, and just submit the final inference kernel. \n\nIs this a valid approach? Can I use as many kernels as I want and just submit the last one?",
    "918876": "You can use as many kernels for training and experimentation. Your final submitted kernel needs to have the inference pipeline.\n\nNote that preprocessing the test data (and any other processing of test data) will also need to be part of the final kernel because only the submitted kernel can access the entire test data."
  },
  "source": "meta"
}