{
  "id": 304515,
  "title": "Possible with a re-run within a few days the first time?",
  "url": "/competitions/g-research-crypto-forecasting/discussion/304515",
  "author_name": "",
  "post_date": "2022-02-01T17:18:38.401274400Z",
  "votes": 4,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Maybe just wishful thinking and lack of patience, but one must try to ask:</p>\n<p>Would it be possible have the first re-run just within a couple of days? <br>\nAfter all the work it should be interesting to see if the solution works and in what quality 😊</p>",
  "messages": [
    {
      "id": "1671719",
      "postDate": "02/01/2022 17:18:38",
      "content": "<p>Maybe just wishful thinking and lack of patience, but one must try to ask:</p>\n<p>Would it be possible have the first re-run just within a couple of days? <br>\nAfter all the work it should be interesting to see if the solution works and in what quality 😊</p>",
      "rawMarkdown": "Maybe just wishful thinking and lack of patience, but one must try to ask:\n\nWould it be possible have the first re-run just within a couple of days? \nAfter all the work it should be interesting to see if the solution works and in what quality 😊",
      "votes": null
    },
    {
      "id": "1671920",
      "postDate": "02/01/2022 19:41:05",
      "content": "<p>I have found intraday corr to be very volatile. Any rerun within a couple of day will be very volatile.</p>",
      "rawMarkdown": "I have found intraday corr to be very volatile. Any rerun within a couple of day will be very volatile.",
      "votes": null
    },
    {
      "id": "1672220",
      "postDate": "02/01/2022 23:58:02",
      "content": "<p>Even with 90-day test periods, my score average and standard deviation across folds was about .06 +- .015. Based on that, I gather that at the end of the competition we'll still only roughly know good our solutions are. Final placement will depend a lot on luck. A few days would give us almost no information, other than that our kernels ran successfully.</p>",
      "rawMarkdown": "Even with 90-day test periods, my score average and standard deviation across folds was about .06 +- .015. Based on that, I gather that at the end of the competition we'll still only roughly know good our solutions are. Final placement will depend a lot on luck. A few days would give us almost no information, other than that our kernels ran successfully.",
      "votes": null
    },
    {
      "id": "1672266",
      "postDate": "02/02/2022 01:39:21",
      "content": "<p>What was your best score on the public lb without any target leakage?</p>",
      "rawMarkdown": "What was your best score on the public lb without any target leakage?",
      "votes": null
    },
    {
      "id": "1673411",
      "postDate": "02/02/2022 17:54:52",
      "content": "<p>.038, which matched my local CV at the time. (Purely by coincidence, I'm sure.) I don't know how high my final model would have scored without target leakage, since I forgot to try that. I would have only done it out of curiosity, because I think local CV is all that counts for optimization, especially in this competition. My final model scored .075 on the public leaderboard, but that's with target leakage.</p>",
      "rawMarkdown": ".038, which matched my local CV at the time. (Purely by coincidence, I'm sure.) I don't know how high my final model would have scored without target leakage, since I forgot to try that. I would have only done it out of curiosity, because I think local CV is all that counts for optimization, especially in this competition. My final model scored .075 on the public leaderboard, but that's with target leakage.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1671920,
      "author_name": "lucasmorin",
      "author_url": "",
      "post_date": "02/01/2022 19:41:05",
      "content": "<p>I have found intraday corr to be very volatile. Any rerun within a couple of day will be very volatile.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1672220,
      "author_name": "nathanrm",
      "author_url": "",
      "post_date": "02/01/2022 23:58:02",
      "content": "<p>Even with 90-day test periods, my score average and standard deviation across folds was about .06 +- .015. Based on that, I gather that at the end of the competition we'll still only roughly know good our solutions are. Final placement will depend a lot on luck. A few days would give us almost no information, other than that our kernels ran successfully.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1672266,
          "author_name": "petersk20",
          "author_url": "",
          "post_date": "02/02/2022 01:39:21",
          "content": "<p>What was your best score on the public lb without any target leakage?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1673411,
          "author_name": "nathanrm",
          "author_url": "",
          "post_date": "02/02/2022 17:54:52",
          "content": "<p>.038, which matched my local CV at the time. (Purely by coincidence, I'm sure.) I don't know how high my final model would have scored without target leakage, since I forgot to try that. I would have only done it out of curiosity, because I think local CV is all that counts for optimization, especially in this competition. My final model scored .075 on the public leaderboard, but that's with target leakage.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1671719": "Maybe just wishful thinking and lack of patience, but one must try to ask:\n\nWould it be possible have the first re-run just within a couple of days? \nAfter all the work it should be interesting to see if the solution works and in what quality 😊",
    "1671920": "I have found intraday corr to be very volatile. Any rerun within a couple of day will be very volatile.",
    "1672220": "Even with 90-day test periods, my score average and standard deviation across folds was about .06 +- .015. Based on that, I gather that at the end of the competition we'll still only roughly know good our solutions are. Final placement will depend a lot on luck. A few days would give us almost no information, other than that our kernels ran successfully.",
    "1672266": "What was your best score on the public lb without any target leakage?",
    "1673411": ".038, which matched my local CV at the time. (Purely by coincidence, I'm sure.) I don't know how high my final model would have scored without target leakage, since I forgot to try that. I would have only done it out of curiosity, because I think local CV is all that counts for optimization, especially in this competition. My final model scored .075 on the public leaderboard, but that's with target leakage."
  },
  "source": "meta"
}