{
  "id": 291477,
  "title": "Baseline forecast using historic average of target [local CV=0.00188]",
  "url": "/competitions/g-research-crypto-forecasting/discussion/291477",
  "author_name": "Antony Southworth",
  "post_date": "2021-11-29T13:18:39.770000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi all,</p>\n<p>Using a technique similar to bootstrap sampling to estimate the performance of simply using the average value of the target (per asset) as the prediction. \"trained\" on 1 week (chosen arbitrarily), evaluated on next 2 weeks (because new data provided every 2 weeks during submissions) with 15 minute gap to avoid leakage. Also to note, I just use <code>cor</code> R function to calculate correlation, ignoring the weighting.</p>\n<p>I don't use notebooks much, instead can find my analysis here:</p>\n<p><a href=\"https://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html\" target=\"_blank\">https://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html</a></p>\n<p>Code is on GitHub if want to inspect:</p>\n<p><a href=\"https://github.com/KyussCaesar/kaggle-cryptoforecast\" target=\"_blank\">https://github.com/KyussCaesar/kaggle-cryptoforecast</a></p>",
  "messages": [
    {
      "id": 1599479,
      "postDate": "2021-11-29T13:18:39.770Z",
      "content": "<p>Hi all,</p>\n<p>Using a technique similar to bootstrap sampling to estimate the performance of simply using the average value of the target (per asset) as the prediction. \"trained\" on 1 week (chosen arbitrarily), evaluated on next 2 weeks (because new data provided every 2 weeks during submissions) with 15 minute gap to avoid leakage. Also to note, I just use <code>cor</code> R function to calculate correlation, ignoring the weighting.</p>\n<p>I don't use notebooks much, instead can find my analysis here:</p>\n<p><a href=\"https://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html\" target=\"_blank\">https://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html</a></p>\n<p>Code is on GitHub if want to inspect:</p>\n<p><a href=\"https://github.com/KyussCaesar/kaggle-cryptoforecast\" target=\"_blank\">https://github.com/KyussCaesar/kaggle-cryptoforecast</a></p>",
      "rawMarkdown": "Hi all,\n\nUsing a technique similar to bootstrap sampling to estimate the performance of simply using the average value of the target (per asset) as the prediction. \"trained\" on 1 week (chosen arbitrarily), evaluated on next 2 weeks (because new data provided every 2 weeks during submissions) with 15 minute gap to avoid leakage. Also to note, I just use `cor` R function to calculate correlation, ignoring the weighting.\n\nI don't use notebooks much, instead can find my analysis here:\n\nhttps://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html\n\nCode is on GitHub if want to inspect:\n\nhttps://github.com/KyussCaesar/kaggle-cryptoforecast",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1599479": "Hi all,\n\nUsing a technique similar to bootstrap sampling to estimate the performance of simply using the average value of the target (per asset) as the prediction. \"trained\" on 1 week (chosen arbitrarily), evaluated on next 2 weeks (because new data provided every 2 weeks during submissions) with 15 minute gap to avoid leakage. Also to note, I just use `cor` R function to calculate correlation, ignoring the weighting.\n\nI don't use notebooks much, instead can find my analysis here:\n\nhttps://kyusscaesar.github.io/kaggle-cryptoforecast/model-baseline-avg.html\n\nCode is on GitHub if want to inspect:\n\nhttps://github.com/KyussCaesar/kaggle-cryptoforecast"
  }
}