{
  "id": 190625,
  "title": "No one using categorical features?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/190625",
  "author_name": "",
  "post_date": "2020-10-12T17:32:21.556779100Z",
  "votes": 3,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I know that we here are a bit in a confuse about the private test, but as of now I couldn't find any notebook using categorical features like content_id. Did anyone try to do it at least?</p>",
  "messages": [
    {
      "id": "1047531",
      "postDate": "10/12/2020 17:32:21",
      "content": "<p>I know that we here are a bit in a confuse about the private test, but as of now I couldn't find any notebook using categorical features like content_id. Did anyone try to do it at least?</p>",
      "rawMarkdown": "I know that we here are a bit in a confuse about the private test, but as of now I couldn't find any notebook using categorical features like content_id. Did anyone try to do it at least?",
      "votes": null
    },
    {
      "id": "1047543",
      "postDate": "10/12/2020 17:42:50",
      "content": "<p>I have used the 'part' column from the questions data as a categorical feature. It has a lot of information overlap with historical information about a question (let's call it historical question accuracy), so while it does add something, I think it is very limited based on what I have seen.</p>\n<p>I do have a few other ideas for categorical features at the moment, but I haven't tested them and there aren't any I am super optimistic about right now.</p>\n<p>I would be interested to hear what others have tried.</p>",
      "rawMarkdown": "I have used the 'part' column from the questions data as a categorical feature. It has a lot of information overlap with historical information about a question (let's call it historical question accuracy), so while it does add something, I think it is very limited based on what I have seen.\n\nI do have a few other ideas for categorical features at the moment, but I haven't tested them and there aren't any I am super optimistic about right now.\n\nI would be interested to hear what others have tried.",
      "votes": null
    },
    {
      "id": "1047550",
      "postDate": "10/12/2020 17:52:59",
      "content": "<p>Nice, will there be content_id in the test that is not on the train? maybe a dumb question, I have just entered the competition</p>",
      "rawMarkdown": "Nice, will there be content_id in the test that is not on the train? maybe a dumb question, I have just entered the competition",
      "votes": null
    },
    {
      "id": "1047552",
      "postDate": "10/12/2020 17:56:12",
      "content": "<p>Yes, I am fairly certain there will be based on what I have read. Presumably, there will also be a new question dataset to accompany it, but that isn't totally clear.</p>",
      "rawMarkdown": "Yes, I am fairly certain there will be based on what I have read. Presumably, there will also be a new question dataset to accompany it, but that isn't totally clear.",
      "votes": null
    },
    {
      "id": "1047923",
      "postDate": "10/13/2020 04:07:08",
      "content": "<p>The data section states that there might be. However all of the content_ids are present in either questions.csv or lectures.csv, so metadata is always available eventhough historical data might not be available.</p>",
      "rawMarkdown": "The data section states that there might be. However all of the content_ids are present in either questions.csv or lectures.csv, so metadata is always available eventhough historical data might not be available.",
      "votes": null
    },
    {
      "id": "1051106",
      "postDate": "10/16/2020 07:11:27",
      "content": "<p>Now I am using parts and bundle_id features from questions.csv with pytorch entity embedding<br>\n<a href=\"https://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230\" target=\"_blank\">https://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230</a></p>",
      "rawMarkdown": "Now I am using parts and bundle_id features from questions.csv with pytorch entity embedding\nhttps://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1047543,
      "author_name": "dwit392",
      "author_url": "",
      "post_date": "10/12/2020 17:42:50",
      "content": "<p>I have used the 'part' column from the questions data as a categorical feature. It has a lot of information overlap with historical information about a question (let's call it historical question accuracy), so while it does add something, I think it is very limited based on what I have seen.</p>\n<p>I do have a few other ideas for categorical features at the moment, but I haven't tested them and there aren't any I am super optimistic about right now.</p>\n<p>I would be interested to hear what others have tried.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1047550,
          "author_name": "shahules",
          "author_url": "",
          "post_date": "10/12/2020 17:52:59",
          "content": "<p>Nice, will there be content_id in the test that is not on the train? maybe a dumb question, I have just entered the competition</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1047552,
          "author_name": "dwit392",
          "author_url": "",
          "post_date": "10/12/2020 17:56:12",
          "content": "<p>Yes, I am fairly certain there will be based on what I have read. Presumably, there will also be a new question dataset to accompany it, but that isn't totally clear.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1047923,
          "author_name": "abhimanyud",
          "author_url": "",
          "post_date": "10/13/2020 04:07:08",
          "content": "<p>The data section states that there might be. However all of the content_ids are present in either questions.csv or lectures.csv, so metadata is always available eventhough historical data might not be available.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1051106,
      "author_name": "shahules",
      "author_url": "",
      "post_date": "10/16/2020 07:11:27",
      "content": "<p>Now I am using parts and bundle_id features from questions.csv with pytorch entity embedding<br>\n<a href=\"https://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230\" target=\"_blank\">https://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1047531": "I know that we here are a bit in a confuse about the private test, but as of now I couldn't find any notebook using categorical features like content_id. Did anyone try to do it at least?",
    "1047543": "I have used the 'part' column from the questions data as a categorical feature. It has a lot of information overlap with historical information about a question (let's call it historical question accuracy), so while it does add something, I think it is very limited based on what I have seen.\n\nI do have a few other ideas for categorical features at the moment, but I haven't tested them and there aren't any I am super optimistic about right now.\n\nI would be interested to hear what others have tried.",
    "1047550": "Nice, will there be content_id in the test that is not on the train? maybe a dumb question, I have just entered the competition",
    "1047552": "Yes, I am fairly certain there will be based on what I have read. Presumably, there will also be a new question dataset to accompany it, but that isn't totally clear.",
    "1047923": "The data section states that there might be. However all of the content_ids are present in either questions.csv or lectures.csv, so metadata is always available eventhough historical data might not be available.",
    "1051106": "Now I am using parts and bundle_id features from questions.csv with pytorch entity embedding\nhttps://www.kaggle.com/shahules/pytorch-entity-embedding?scriptVersionId=44817230"
  },
  "source": "meta"
}