![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3273905%2F4c409b21afcef0f3e5ae172d6090a844%2FWX20201212-0644132x.png?generation=1607726695891832&alt=media\")
Hi @sohier and organizers,
\nI had a query related to the questions metadata. The question is as follows:
\nIn train.csv we have a column 'task_container_id'. The description of this column says \"code for the batch of questions or lectures. For example, a user might see three questions in a row before seeing the explanations for any of them. Those three would all share a task_container_id\"
\nNow, the questions.csv has a column 'bundle_id'. The description of this column says \"code for which questions are served together\".
\nThus, for rows in train/test where the content type is question (0), would it be fair to assume that 'task_container_id' is the same as 'bundle_id'?
\nI am entering this competition just now, so I apologize if this is a repeat question from some other thread.
\nCheers!!
", "messages": [ { "id": "1048438", "postDate": "10/13/2020 13:51:00", "content": "Hi @sohier and organizers,
\nI had a query related to the questions metadata. The question is as follows:
\nIn train.csv we have a column 'task_container_id'. The description of this column says \"code for the batch of questions or lectures. For example, a user might see three questions in a row before seeing the explanations for any of them. Those three would all share a task_container_id\"
\nNow, the questions.csv has a column 'bundle_id'. The description of this column says \"code for which questions are served together\".
\nThus, for rows in train/test where the content type is question (0), would it be fair to assume that 'task_container_id' is the same as 'bundle_id'?
\nI am entering this competition just now, so I apologize if this is a repeat question from some other thread.
\nCheers!!
", "rawMarkdown": "Hi @sohier and organizers,\n\nI had a query related to the questions metadata. The question is as follows:\n\nIn train.csv we have a column 'task_container_id'. The description of this column says \"code for the batch of questions or lectures. For example, a user might see three questions in a row before seeing the explanations for any of them. Those three would all share a task_container_id\"\n\nNow, the questions.csv has a column 'bundle_id'. The description of this column says \"code for which questions are served together\".\n\nThus, for rows in train/test where the content type is question (0), would it be fair to assume that 'task_container_id' is the same as 'bundle_id'?\n\nI am entering this competition just now, so I apologize if this is a repeat question from some other thread.\n\nCheers!!", "votes": null }, { "id": "1051204", "postDate": "10/16/2020 08:59:45", "content": "@sandy1112 I had the very same question in mind. Although the description in questions.csv says so, I think it's much more reliable to use task_container_id instead of bundle_id.
Reason:
\nprior_question_elapsed_time is the avg time a student takes to complete the previous batch (total_time taken / total_no_questions). So it should follow that prior_question_elapsed time must be the same for any row of any user for particular bundle (bundle being a group of questions that a user sees at a time).
I tried verifying the above statement by grouping all rows and checking the number of unique prior_question_elapsed_time values:
\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.groupby(['user_id', 'task_container_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> True\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.merge(ql, on=['content_id', 'content_type_id'])\n .groupby(['user_id', 'bundle_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> False\nMy logic is that since the other features match better with the assumption that each bundle have the same task_container_id, it is much better to use this feature instead of using bundle_id. I may be totally wrong. Do let me know your thoughts on this.
", "rawMarkdown": "sandy1112 I had the very same question in mind. Although the description in questions.csv says so, I think it's much more reliable to use `task_container_id` instead of `bundle_id`. \n\nReason:\n`prior_question_elapsed_time` is the avg time a student takes to complete the previous batch (total_time taken / total_no_questions). So it should follow that prior_question_elapsed time must be the same for any row of any user for particular *bundle* (bundle being a group of questions that a user sees at a time). \n\nI tried verifying the above statement by grouping all rows and checking the number of unique prior_question_elapsed_time values:\n\n- Grouping by ['user_id', 'task_container_id']:\n\n```python3\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.groupby(['user_id', 'task_container_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> True\n```\n\n- Grouping by ['user_id', 'bundle_id']\n\n```python3\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.merge(ql, on=['content_id', 'content_type_id'])\n .groupby(['user_id', 'bundle_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> False\n```\n\nMy logic is that since the other features match better with the assumption that each bundle have the same task_container_id, it is much better to use this feature instead of using bundle_id. I may be totally wrong. Do let me know your thoughts on this.", "votes": null }, { "id": "1051526", "postDate": "10/16/2020 15:50:19", "content": "They're totally distinct concepts. Questions with the same bundle ID should be served together, but every time they are served they will be under a different task container id (for a given user).
", "rawMarkdown": "They're totally distinct concepts. Questions with the same bundle ID should be served together, but every time they are served they will be under a different task container id (for a given user).", "votes": null }, { "id": "1051533", "postDate": "10/16/2020 15:54:49", "content": "Thanks for confirming. Appreciate the help on this.
", "rawMarkdown": "Thanks for confirming. Appreciate the help on this.", "votes": null }, { "id": "1064479", "postDate": "10/30/2020 07:21:45", "content": "@sohier does it means that bundle id could remain same for set of questions grouped under one.. but they may keep appering under different Task container ids together as bundle
", "rawMarkdown": "sohier does it means that bundle id could remain same for set of questions grouped under one.. but they may keep appering under different Task container ids together as bundle", "votes": null }, { "id": "1109640", "postDate": "12/11/2020 22:45:53", "content": "@sohier how to explain same bunld_id with same contrainer_id and same user_id?
\n
\n\neach bundle have the same task_container_id
\n
Does this mean, even for different users, if the questions are from the same bundle, these questions share the same task_container_id?
Hi @wuwenmin, questions of the same bundle share the same task_container_id but for different users the shared task_container_ids might still be different. task_container_id's purpose is to point at the order in which the user had seen the questions. It is not guaranteed to be same for different users.
For one user the question bundle containing 3 questions might have task_container_ids as [2, 2, 2] whereas for another user the same question bundle might be something like [998, 998, 998]. Hope this helps :)
", "rawMarkdown": "Hi @wuwenmin, questions of the same bundle share the same `task_container_id` but for different users the *shared* task_container_ids might still be different. `task_container_id`'s purpose is to point at the order in which the user had seen the questions. It is not guaranteed to be *same* for different users. \n\nFor one user the question bundle containing 3 questions might have task_container_ids as [2, 2, 2] whereas for another user the same question bundle might be something like [998, 998, 998]. Hope this helps :)", "votes": null }, { "id": "1118390", "postDate": "12/19/2020 03:01:43", "content": "Thanks for your reply. Yes, I just checked the data, the following are the distribution of the number of unique questions of the container task id:
\ncount 10000.00000\nmean 2766.53760\nstd 3155.41549\nmin 165.00000\n25% 411.75000\n50% 1228.00000\n75% 4263.50000\nmax 11503.00000\nName: num_ques, dtype: float64\nI saw some public notebooks use the task_container_id as a feature, so I was wondering whether the task_container_id indicates the group of questions for all users. It's not.
@sandy1112 I had the very same question in mind. Although the description in questions.csv says so, I think it's much more reliable to use task_container_id instead of bundle_id.
Reason:
\nprior_question_elapsed_time is the avg time a student takes to complete the previous batch (total_time taken / total_no_questions). So it should follow that prior_question_elapsed time must be the same for any row of any user for particular bundle (bundle being a group of questions that a user sees at a time).
I tried verifying the above statement by grouping all rows and checking the number of unique prior_question_elapsed_time values:
\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.groupby(['user_id', 'task_container_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> True\n# unique prior_elapsed_time per bundle is always one (0 for nans)\n(data.merge(ql, on=['content_id', 'content_type_id'])\n .groupby(['user_id', 'bundle_id'])\n ['prior_question_elapsed_time']\n .nunique().values <= 1).all()\n\n>>> False\nMy logic is that since the other features match better with the assumption that each bundle have the same task_container_id, it is much better to use this feature instead of using bundle_id. I may be totally wrong. Do let me know your thoughts on this.
", "votes": null, "replies": [ { "id": 1118325, "author_name": "wuwenmin", "author_url": "", "post_date": "12/19/2020 00:20:32", "content": "\n\neach bundle have the same task_container_id
\n
Does this mean, even for different users, if the questions are from the same bundle, these questions share the same task_container_id?
Hi @wuwenmin, questions of the same bundle share the same task_container_id but for different users the shared task_container_ids might still be different. task_container_id's purpose is to point at the order in which the user had seen the questions. It is not guaranteed to be same for different users.
For one user the question bundle containing 3 questions might have task_container_ids as [2, 2, 2] whereas for another user the same question bundle might be something like [998, 998, 998]. Hope this helps :)
", "votes": null, "replies": [] }, { "id": 1118390, "author_name": "wuwenmin", "author_url": "", "post_date": "12/19/2020 03:01:43", "content": "Thanks for your reply. Yes, I just checked the data, the following are the distribution of the number of unique questions of the container task id:
\ncount 10000.00000\nmean 2766.53760\nstd 3155.41549\nmin 165.00000\n25% 411.75000\n50% 1228.00000\n75% 4263.50000\nmax 11503.00000\nName: num_ques, dtype: float64\nI saw some public notebooks use the task_container_id as a feature, so I was wondering whether the task_container_id indicates the group of questions for all users. It's not.
They're totally distinct concepts. Questions with the same bundle ID should be served together, but every time they are served they will be under a different task container id (for a given user).
", "votes": null, "replies": [ { "id": 1051533, "author_name": "sandy1112", "author_url": "", "post_date": "10/16/2020 15:54:49", "content": "Thanks for confirming. Appreciate the help on this.
", "votes": null, "replies": [] }, { "id": 1064479, "author_name": "jaideepvalani", "author_url": "", "post_date": "10/30/2020 07:21:45", "content": "@sohier does it means that bundle id could remain same for set of questions grouped under one.. but they may keep appering under different Task container ids together as bundle
", "votes": null, "replies": [] }, { "id": 1109640, "author_name": "jt120lz", "author_url": "", "post_date": "12/11/2020 22:45:53", "content": "@sohier how to explain same bunld_id with same contrainer_id and same user_id?
\n