{
  "id": 587949,
  "title": "What kind of external data can we possibly use?",
  "url": "/competitions/drw-crypto-market-prediction/discussion/587949",
  "author_name": "",
  "post_date": "2025-07-03T15:17:59.926620700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>In the intro it says 'You’re welcome to use all the data we provide, as well as any freely and publicly available external datasets', but any external data we used for train would not be possible to be used for inference isn't it? The timestamp are masked &amp; shuffled so I wonder what kind of external data can be added both in train and test?</p>",
  "messages": [
    {
      "id": "3240207",
      "postDate": "07/03/2025 15:17:59",
      "content": "<p>In the intro it says 'You’re welcome to use all the data we provide, as well as any freely and publicly available external datasets', but any external data we used for train would not be possible to be used for inference isn't it? The timestamp are masked &amp; shuffled so I wonder what kind of external data can be added both in train and test?</p>",
      "rawMarkdown": "In the intro it says 'You’re welcome to use all the data we provide, as well as any freely and publicly available external datasets', but any external data we used for train would not be possible to be used for inference isn't it? The timestamp are masked & shuffled so I wonder what kind of external data can be added both in train and test?",
      "votes": null
    },
    {
      "id": "3240222",
      "postDate": "07/03/2025 15:29:30",
      "content": "<p>I suppose you could find external crypto data with the the minimum market features and train a model on that and then see how that model compares to a model trained on the DRW crypto data. It isn't a perfect science but you might be able to learn a few things or create new assumptions to test.</p>\n<p>You could even try to better understand if one of the proprietary variables closely matches a publicly available feature; if a decent match is found you would have a better understanding of what a proprietary variable resembles and this could inform feature engineering.</p>\n<p>I also had the idea of looking for any open source crypto models that work with market features and see if that could be used as an analysis tool.</p>",
      "rawMarkdown": "I suppose you could find external crypto data with the the minimum market features and train a model on that and then see how that model compares to a model trained on the DRW crypto data. It isn't a perfect science but you might be able to learn a few things or create new assumptions to test.\n\nYou could even try to better understand if one of the proprietary variables closely matches a publicly available feature; if a decent match is found you would have a better understanding of what a proprietary variable resembles and this could inform feature engineering.\n\nI also had the idea of looking for any open source crypto models that work with market features and see if that could be used as an analysis tool.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3240222,
      "author_name": "taylorsamarel",
      "author_url": "",
      "post_date": "07/03/2025 15:29:30",
      "content": "<p>I suppose you could find external crypto data with the the minimum market features and train a model on that and then see how that model compares to a model trained on the DRW crypto data. It isn't a perfect science but you might be able to learn a few things or create new assumptions to test.</p>\n<p>You could even try to better understand if one of the proprietary variables closely matches a publicly available feature; if a decent match is found you would have a better understanding of what a proprietary variable resembles and this could inform feature engineering.</p>\n<p>I also had the idea of looking for any open source crypto models that work with market features and see if that could be used as an analysis tool.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3240207": "In the intro it says 'You’re welcome to use all the data we provide, as well as any freely and publicly available external datasets', but any external data we used for train would not be possible to be used for inference isn't it? The timestamp are masked & shuffled so I wonder what kind of external data can be added both in train and test?",
    "3240222": "I suppose you could find external crypto data with the the minimum market features and train a model on that and then see how that model compares to a model trained on the DRW crypto data. It isn't a perfect science but you might be able to learn a few things or create new assumptions to test.\n\nYou could even try to better understand if one of the proprietary variables closely matches a publicly available feature; if a decent match is found you would have a better understanding of what a proprietary variable resembles and this could inform feature engineering.\n\nI also had the idea of looking for any open source crypto models that work with market features and see if that could be used as an analysis tool."
  },
  "source": "meta"
}