![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F449263%2F2fa4e9baa35cef6de528bf5c1470acb9%2Fprior_question_elapsed_time.png?generation=1605269696873806&alt=media\")
My understanding so far is:
\n(See discussions like: https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189351)
timestamp: time when action ended - with end of first action being the 0 mark
\nprior_question_elapsed_time: Average time per question of LAST question/set of questions.
If this is the case:
\nHow can prior_question_elapsed_time be not NaN for first action of some users?
\n
Not sure if I am missing anything here? @sohier
", "messages": [ { "id": "1077274", "postDate": "11/13/2020 12:16:37", "content": "My understanding so far is:
\n(See discussions like: https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189351)
timestamp: time when action ended - with end of first action being the 0 mark
\nprior_question_elapsed_time: Average time per question of LAST question/set of questions.
If this is the case:
\nHow can prior_question_elapsed_time be not NaN for first action of some users?
\n
Not sure if I am missing anything here? @sohier
", "rawMarkdown": "My understanding so far is:\n(See discussions like: https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189351)\n\ntimestamp: time when action ended - with end of first action being the 0 mark\nprior_question_elapsed_time: Average time per question of LAST question/set of questions.\n\nIf this is the case: \n\nHow can prior_question_elapsed_time be not NaN for first action of some users?\n\n\n\nNot sure if I am missing anything here? @sohier", "votes": null }, { "id": "1077280", "postDate": "11/13/2020 12:22:25", "content": "Given that the value for first two task containers is the same (just quick checked on this one sample above) maybe there was a problem in data preparation that the NaN was not set correctly?
", "rawMarkdown": "Given that the value for first two task containers is the same (just quick checked on this one sample above) maybe there was a problem in data preparation that the NaN was not set correctly?", "votes": null }, { "id": "1077306", "postDate": "11/13/2020 12:57:53", "content": "Good find! I checked it as well. Seems like it happens when the first questions are a bundle (share same task container id). Should be NaN I suppose. But the value is the same as the value of the next task container id - like you mentioned.
", "rawMarkdown": "Good find! I checked it as well. Seems like it happens when the first questions are a bundle (share same task container id). Should be NaN I suppose. But the value is the same as the value of the next task container id - like you mentioned.", "votes": null }, { "id": "1077341", "postDate": "11/13/2020 13:44:12", "content": "I was about to suggest that a dirty-hack-quick-fix might be something like:
\ntrain.loc[train.task_container_id == 0, 'prior_question_elapsed_time'] = VALUE
but then I realized that for example user 1752395680 starts with task_container_id 1 for lecture and 2 for questions …
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F449263%2F6542a5b15829925d5cc934dae25ad822%2Fprior_question_elapsed_time_v02.png?generation=1605274939955491&alt=media\")
Anybody understanding this? @sohier
\nPS: My assumption of course is that the train/test split in this competition is respecting time….
", "rawMarkdown": "I was about to suggest that a dirty-hack-quick-fix might be something like:\n\n`train.loc[train.task_container_id == 0, 'prior_question_elapsed_time'] = VALUE`\n\nbut then I realized that for example user 1752395680 starts with task_container_id 1 for lecture and 2 for questions ...\n\n\n\nAnybody understanding this? @sohier \n\nPS: My assumption of course is that the train/test split in this competition is respecting time....", "votes": null }, { "id": "1077366", "postDate": "11/13/2020 14:22:27", "content": "Well there's no assurance that you have the data right from a user's history starting, right?
", "rawMarkdown": "Well there's no assurance that you have the data right from a user's history starting, right?", "votes": null }, { "id": "1077369", "postDate": "11/13/2020 14:23:59", "content": "Then I would expect a non zero timestamp for consistency?
", "rawMarkdown": "Then I would expect a non zero timestamp for consistency?", "votes": null }, { "id": "1077875", "postDate": "11/14/2020 03:26:57", "content": "This is the same questions with mine (2 posts below): https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/196879! I still don't understand why it has not NaN value even though they have timestamp of 0….
@matthiasanderer
\nto ensure prior time refers to last q attempted in sequence while training ,relying on timestamp order is sufficient ?
Given that the value for first two task containers is the same (just quick checked on this one sample above) maybe there was a problem in data preparation that the NaN was not set correctly?
", "votes": null, "replies": [] }, { "id": 1077306, "author_name": "rafiko1", "author_url": "", "post_date": "11/13/2020 12:57:53", "content": "Good find! I checked it as well. Seems like it happens when the first questions are a bundle (share same task container id). Should be NaN I suppose. But the value is the same as the value of the next task container id - like you mentioned.
", "votes": null, "replies": [] }, { "id": 1077341, "author_name": "matthiasanderer", "author_url": "", "post_date": "11/13/2020 13:44:12", "content": "I was about to suggest that a dirty-hack-quick-fix might be something like:
\ntrain.loc[train.task_container_id == 0, 'prior_question_elapsed_time'] = VALUE
but then I realized that for example user 1752395680 starts with task_container_id 1 for lecture and 2 for questions …
\n
Anybody understanding this? @sohier
\nPS: My assumption of course is that the train/test split in this competition is respecting time….
", "votes": null, "replies": [] }, { "id": 1077366, "author_name": "adityaecdrid", "author_url": "", "post_date": "11/13/2020 14:22:27", "content": "Well there's no assurance that you have the data right from a user's history starting, right?
", "votes": null, "replies": [ { "id": 1077369, "author_name": "matthiasanderer", "author_url": "", "post_date": "11/13/2020 14:23:59", "content": "Then I would expect a non zero timestamp for consistency?
", "votes": null, "replies": [] } ] }, { "id": 1077875, "author_name": "choisgood", "author_url": "", "post_date": "11/14/2020 03:26:57", "content": "This is the same questions with mine (2 posts below): https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/196879! I still don't understand why it has not NaN value even though they have timestamp of 0….
@matthiasanderer
\nto ensure prior time refers to last q attempted in sequence while training ,relying on timestamp order is sufficient ?