{
  "id": 303886,
  "title": "Inconsistency on submission run times",
  "url": "/competitions/g-research-crypto-forecasting/discussion/303886",
  "author_name": "",
  "post_date": "2022-01-29T20:53:00.917263700Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I've had some notebooks that run in 700seconds that made it past 9 hour submission limit and now some that run in 600s that are failing..</p>\n<p>Anyone have ideas what might be causing a notebook on submission stage to differ so much? Can we get some clarity or guidance on the 9 hour limit.. if we can pass that requirement now, are safe in the next 2-3 months of forward testing? </p>\n<p>Going to be a lot of notebook time outs if they are variable in the compute run times..</p>",
  "messages": [
    {
      "id": "1668646",
      "postDate": "01/29/2022 20:53:00",
      "content": "<p>I've had some notebooks that run in 700seconds that made it past 9 hour submission limit and now some that run in 600s that are failing..</p>\n<p>Anyone have ideas what might be causing a notebook on submission stage to differ so much? Can we get some clarity or guidance on the 9 hour limit.. if we can pass that requirement now, are safe in the next 2-3 months of forward testing? </p>\n<p>Going to be a lot of notebook time outs if they are variable in the compute run times..</p>",
      "rawMarkdown": "I've had some notebooks that run in 700seconds that made it past 9 hour submission limit and now some that run in 600s that are failing..\n\nAnyone have ideas what might be causing a notebook on submission stage to differ so much? Can we get some clarity or guidance on the 9 hour limit.. if we can pass that requirement now, are safe in the next 2-3 months of forward testing? \n\nGoing to be a lot of notebook time outs if they are variable in the compute run times..",
      "votes": null
    },
    {
      "id": "1668695",
      "postDate": "01/29/2022 23:28:11",
      "content": "<p>I've noticed this too and it'd be a real shame for kernels to fail at the last leaderboard update (which is the only update when the full test data is run). </p>\n<p>Best to build in some buffer time. Either that or you could periodically keep track of how long your submission has taken so far, and if you only have 15 minutes left, switch everything off and make random or zero predictions which are very fast.</p>",
      "rawMarkdown": "I've noticed this too and it'd be a real shame for kernels to fail at the last leaderboard update (which is the only update when the full test data is run). \n\nBest to build in some buffer time. Either that or you could periodically keep track of how long your submission has taken so far, and if you only have 15 minutes left, switch everything off and make random or zero predictions which are very fast.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1668695,
      "author_name": "jagofc",
      "author_url": "",
      "post_date": "01/29/2022 23:28:11",
      "content": "<p>I've noticed this too and it'd be a real shame for kernels to fail at the last leaderboard update (which is the only update when the full test data is run). </p>\n<p>Best to build in some buffer time. Either that or you could periodically keep track of how long your submission has taken so far, and if you only have 15 minutes left, switch everything off and make random or zero predictions which are very fast.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1668646": "I've had some notebooks that run in 700seconds that made it past 9 hour submission limit and now some that run in 600s that are failing..\n\nAnyone have ideas what might be causing a notebook on submission stage to differ so much? Can we get some clarity or guidance on the 9 hour limit.. if we can pass that requirement now, are safe in the next 2-3 months of forward testing? \n\nGoing to be a lot of notebook time outs if they are variable in the compute run times..",
    "1668695": "I've noticed this too and it'd be a real shame for kernels to fail at the last leaderboard update (which is the only update when the full test data is run). \n\nBest to build in some buffer time. Either that or you could periodically keep track of how long your submission has taken so far, and if you only have 15 minutes left, switch everything off and make random or zero predictions which are very fast."
  },
  "source": "meta"
}