{
  "id": 580557,
  "title": "Is it possible to implement Time Series based models?",
  "url": "/competitions/drw-crypto-market-prediction/discussion/580557",
  "author_name": "",
  "post_date": "2025-05-24T23:15:27.740381100Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Since Timestamps are not given, it looks like pure regression task rather than time series forecasting. </p>\n<p>The Host said \"timestamps are masked\", so i am not sure if it starts directly from 1st march, 2024 or later. also it's not mentioned that it's shuffled or not but i am assuming it's shuffled. Please let me know if it's already in order?</p>\n<p>By plotting the label from training data, it is cleat that underlying asset is BTC from March 1, 2023, to February 29, 2024. So is there any way to figure out correct order of test data? is there any feature which directly resembles price movement through which we can arrange the order of the test set.</p>\n<p>Also the host said we are allowed to use external datasets. but how is it possible? cause i don't know any api which freely provides such small timeframe data, even if there are some, it will be only 5-10 features i guess.</p>\n<p>So I am assuming that this is purely about data processing, feature engineering and experimenting with all kind of models and ensembles. Please guide me if i am wrong.</p>",
  "messages": [
    {
      "id": "3208917",
      "postDate": "05/24/2025 23:15:27",
      "content": "<p>Since Timestamps are not given, it looks like pure regression task rather than time series forecasting. </p>\n<p>The Host said \"timestamps are masked\", so i am not sure if it starts directly from 1st march, 2024 or later. also it's not mentioned that it's shuffled or not but i am assuming it's shuffled. Please let me know if it's already in order?</p>\n<p>By plotting the label from training data, it is cleat that underlying asset is BTC from March 1, 2023, to February 29, 2024. So is there any way to figure out correct order of test data? is there any feature which directly resembles price movement through which we can arrange the order of the test set.</p>\n<p>Also the host said we are allowed to use external datasets. but how is it possible? cause i don't know any api which freely provides such small timeframe data, even if there are some, it will be only 5-10 features i guess.</p>\n<p>So I am assuming that this is purely about data processing, feature engineering and experimenting with all kind of models and ensembles. Please guide me if i am wrong.</p>",
      "rawMarkdown": "Since Timestamps are not given, it looks like pure regression task rather than time series forecasting. \n\nThe Host said \"timestamps are masked\", so i am not sure if it starts directly from 1st march, 2024 or later. also it's not mentioned that it's shuffled or not but i am assuming it's shuffled. Please let me know if it's already in order?\n\nBy plotting the label from training data, it is cleat that underlying asset is BTC from March 1, 2023, to February 29, 2024. So is there any way to figure out correct order of test data? is there any feature which directly resembles price movement through which we can arrange the order of the test set.\n\nAlso the host said we are allowed to use external datasets. but how is it possible? cause i don't know any api which freely provides such small timeframe data, even if there are some, it will be only 5-10 features i guess.\n\nSo I am assuming that this is purely about data processing, feature engineering and experimenting with all kind of models and ensembles. Please guide me if i am wrong.",
      "votes": null
    },
    {
      "id": "3209004",
      "postDate": "05/25/2025 05:03:11",
      "content": "<p>I think it's impossible to implement ts models but we might conduct some time series analysis on label to see what we can figure out.</p>\n<p>The test data should start directly from 1st march, 2024 and it's shuffled.</p>",
      "rawMarkdown": "I think it's impossible to implement ts models but we might conduct some time series analysis on label to see what we can figure out.\n\nThe test data should start directly from 1st march, 2024 and it's shuffled.",
      "votes": null
    },
    {
      "id": "3209742",
      "postDate": "05/26/2025 08:52:43",
      "content": "<p>well maybe I have an idea to transfer timestamp into dayofyear and minuteinday then 1 model is to predict those for test data then use it for meta model which include both those 2 and other data. that's what i come up with so it can capture temporal pattern maybe havent tried yet .</p>",
      "rawMarkdown": "well maybe I have an idea to transfer timestamp into dayofyear and minuteinday then 1 model is to predict those for test data then use it for meta model which include both those 2 and other data. that's what i come up with so it can capture temporal pattern maybe havent tried yet .",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3209004,
      "author_name": "yw2735",
      "author_url": "",
      "post_date": "05/25/2025 05:03:11",
      "content": "<p>I think it's impossible to implement ts models but we might conduct some time series analysis on label to see what we can figure out.</p>\n<p>The test data should start directly from 1st march, 2024 and it's shuffled.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3209742,
      "author_name": "theonetheonly",
      "author_url": "",
      "post_date": "05/26/2025 08:52:43",
      "content": "<p>well maybe I have an idea to transfer timestamp into dayofyear and minuteinday then 1 model is to predict those for test data then use it for meta model which include both those 2 and other data. that's what i come up with so it can capture temporal pattern maybe havent tried yet .</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3208917": "Since Timestamps are not given, it looks like pure regression task rather than time series forecasting. \n\nThe Host said \"timestamps are masked\", so i am not sure if it starts directly from 1st march, 2024 or later. also it's not mentioned that it's shuffled or not but i am assuming it's shuffled. Please let me know if it's already in order?\n\nBy plotting the label from training data, it is cleat that underlying asset is BTC from March 1, 2023, to February 29, 2024. So is there any way to figure out correct order of test data? is there any feature which directly resembles price movement through which we can arrange the order of the test set.\n\nAlso the host said we are allowed to use external datasets. but how is it possible? cause i don't know any api which freely provides such small timeframe data, even if there are some, it will be only 5-10 features i guess.\n\nSo I am assuming that this is purely about data processing, feature engineering and experimenting with all kind of models and ensembles. Please guide me if i am wrong.",
    "3209004": "I think it's impossible to implement ts models but we might conduct some time series analysis on label to see what we can figure out.\n\nThe test data should start directly from 1st march, 2024 and it's shuffled.",
    "3209742": "well maybe I have an idea to transfer timestamp into dayofyear and minuteinday then 1 model is to predict those for test data then use it for meta model which include both those 2 and other data. that's what i come up with so it can capture temporal pattern maybe havent tried yet ."
  },
  "source": "meta"
}