{
  "id": 296019,
  "title": "Asset weights change over time?",
  "url": "/competitions/g-research-crypto-forecasting/discussion/296019",
  "author_name": "",
  "post_date": "2021-12-19T11:06:50.527846Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey guys,</p>\n<p>I have a question regarding the weights calculated for each asset. These are described as being based on the capitalisation of each crypto at some point in the past. </p>\n<p>I understand that the weights are given for metric calculation but I am sure we can use it train the model as well. Given the fact that the competition is 3 months long and further 3 months of prediction based on realtime data, do you the weights will have to change after sometime given that market is so volatile? And would you think it work be writing a weights adjusting function that will help make better predictions?</p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": "1622979",
      "postDate": "12/19/2021 11:06:50",
      "content": "<p>Hey guys,</p>\n<p>I have a question regarding the weights calculated for each asset. These are described as being based on the capitalisation of each crypto at some point in the past. </p>\n<p>I understand that the weights are given for metric calculation but I am sure we can use it train the model as well. Given the fact that the competition is 3 months long and further 3 months of prediction based on realtime data, do you the weights will have to change after sometime given that market is so volatile? And would you think it work be writing a weights adjusting function that will help make better predictions?</p>\n<p>Thanks</p>",
      "rawMarkdown": "Hey guys,\n\nI have a question regarding the weights calculated for each asset. These are described as being based on the capitalisation of each crypto at some point in the past. \n\nI understand that the weights are given for metric calculation but I am sure we can use it train the model as well. Given the fact that the competition is 3 months long and further 3 months of prediction based on realtime data, do you the weights will have to change after sometime given that market is so volatile? And would you think it work be writing a weights adjusting function that will help make better predictions?\n\nThanks",
      "votes": null
    },
    {
      "id": "1622996",
      "postDate": "12/19/2021 11:29:07",
      "content": "<p>Since the weights are key to forming the evaluation metric, it would be very unfair if that were to be changed on the fly (- literally changing the rules of the game)! So it should be safe to assume that the weights remain constant.</p>\n<p>Weight-adjusted losses could definitely be valuable if done properly - and not necessarily for making better predictions across all targets, but for making better predictions for the targets which are more heavily weighted in the evaluation metric.</p>",
      "rawMarkdown": "Since the weights are key to forming the evaluation metric, it would be very unfair if that were to be changed on the fly (- literally changing the rules of the game)! So it should be safe to assume that the weights remain constant.\n\nWeight-adjusted losses could definitely be valuable if done properly - and not necessarily for making better predictions across all targets, but for making better predictions for the targets which are more heavily weighted in the evaluation metric.",
      "votes": null
    },
    {
      "id": "1623867",
      "postDate": "12/20/2021 10:54:07",
      "content": "<p>That totally makes sense <a href=\"https://www.kaggle.com/jagofc\" target=\"_blank\">@jagofc</a>. Weight-adjusted losses is something that could help the model make better predictions. </p>",
      "rawMarkdown": "That totally makes sense @jagofc. Weight-adjusted losses is something that could help the model make better predictions.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1622996,
      "author_name": "jagofc",
      "author_url": "",
      "post_date": "12/19/2021 11:29:07",
      "content": "<p>Since the weights are key to forming the evaluation metric, it would be very unfair if that were to be changed on the fly (- literally changing the rules of the game)! So it should be safe to assume that the weights remain constant.</p>\n<p>Weight-adjusted losses could definitely be valuable if done properly - and not necessarily for making better predictions across all targets, but for making better predictions for the targets which are more heavily weighted in the evaluation metric.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1623867,
          "author_name": "nitishmalagi",
          "author_url": "",
          "post_date": "12/20/2021 10:54:07",
          "content": "<p>That totally makes sense <a href=\"https://www.kaggle.com/jagofc\" target=\"_blank\">@jagofc</a>. Weight-adjusted losses is something that could help the model make better predictions. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1622979": "Hey guys,\n\nI have a question regarding the weights calculated for each asset. These are described as being based on the capitalisation of each crypto at some point in the past. \n\nI understand that the weights are given for metric calculation but I am sure we can use it train the model as well. Given the fact that the competition is 3 months long and further 3 months of prediction based on realtime data, do you the weights will have to change after sometime given that market is so volatile? And would you think it work be writing a weights adjusting function that will help make better predictions?\n\nThanks",
    "1622996": "Since the weights are key to forming the evaluation metric, it would be very unfair if that were to be changed on the fly (- literally changing the rules of the game)! So it should be safe to assume that the weights remain constant.\n\nWeight-adjusted losses could definitely be valuable if done properly - and not necessarily for making better predictions across all targets, but for making better predictions for the targets which are more heavily weighted in the evaluation metric.",
    "1623867": "That totally makes sense @jagofc. Weight-adjusted losses is something that could help the model make better predictions."
  },
  "source": "meta"
}