{ "id": 200919, "title": "User Id and TImestamp", "url": "/competitions/riiid-test-answer-prediction/discussion/200919", "author_name": "Jaideep", "post_date": "2020-12-02T11:44:55.153000", "votes": 4, "comment_count": 8, "views": 0, "content": "

Timestamp =0 means if m correct ,time from which the user interaction clock starts .

\n

However there are some doubts i have
\nEach user should have Timestamp staring at 0 . so number of Timestam=0 should be come as Number of unique users.
\nBut i think that is not the case

\n
train_df.user_id.nunique()\n123365\n
\n
train_df[train_df.timestamp==0]\n124586 rows × 9 columns\n
", "messages": [ { "id": 1099473, "postDate": "2020-12-02T11:44:55.153Z", "content": "

Timestamp =0 means if m correct ,time from which the user interaction clock starts .

\n

However there are some doubts i have
\nEach user should have Timestamp staring at 0 . so number of Timestam=0 should be come as Number of unique users.
\nBut i think that is not the case

\n
train_df.user_id.nunique()\n123365\n
\n
train_df[train_df.timestamp==0]\n124586 rows × 9 columns\n
", "rawMarkdown": "Timestamp =0 means if m correct ,time from which the user interaction clock starts . \n\nHowever there are some doubts i have\nEach user should have Timestamp staring at 0 . so number of Timestam=0 should be come as Number of unique users. \nBut i think that is not the case\n```\ntrain_df.user_id.nunique()\n123365\n```\n\n```\ntrain_df[train_df.timestamp==0]\n124586 rows × 9 columns\n```", "votes": 4 }, { "id": 1099622, "postDate": "2020-12-02T13:37:13.497Z", "content": "

As discussed here, timestamp is the time when a task (= unique task_container_id) is completed by a user. So train_df[train_df.timestamp==0] selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.
\nBy the way, we don't always have task_container_id==0 when timestamp==0, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).

", "rawMarkdown": "As discussed [here](https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189465), timestamp is the time when a task (= unique task_container_id) is completed by a user. So `train_df[train_df.timestamp==0]` selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.\nBy the way, we don't always have `task_container_id==0` when `timestamp==0`, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).", "votes": 1, "replies": [ { "id": 1099690, "postDate": "2020-12-02T14:35:54.653Z", "content": "

@johannhuber i checked the discussion it talks on Task container id order more .
\nHere the doubt is why Not all User ids have Timestamp starting 0 , so in next task or interaction we see the time user took to complete the set of previous tasks..

", "rawMarkdown": "@johannhuber i checked the discussion it talks on Task container id order more .\nHere the doubt is why Not all User ids have Timestamp starting 0 , so in next task or interaction we see the time user took to complete the set of previous tasks.. " }, { "id": 1099701, "postDate": "2020-12-02T14:47:43.387Z", "content": "

Then I'm not sure to understand your doubt : what do you mean by \"Not all User ids have Timestamp starting 0\" ?
\nActually I got :
\ntrain.groupby(['user_id'])['timestamp'].first().unique() # array([0])
\ntrain.groupby(['user_id'])['timestamp'].min().unique() # array([0])

", "rawMarkdown": "Then I'm not sure to understand your doubt : what do you mean by \"Not all User ids have Timestamp starting 0\" ?\nActually I got :\n`train.groupby(['user_id'])['timestamp'].first().unique() # array([0])`\n`train.groupby(['user_id'])['timestamp'].min().unique() # array([0])`\n", "votes": 1 }, { "id": 1099714, "postDate": "2020-12-02T14:58:44.817Z", "content": "

you can check my query above for DF.. number of unique users , number rows with Timestamp==0 should come as same, please see if you getting it same

", "rawMarkdown": "you can check my query above for DF.. number of unique users , number rows with Timestamp==0 should come as same, please see if you getting it same" }, { "id": 1099836, "postDate": "2020-12-02T16:13:31.733Z", "content": "

If I understand you well, you're assuming :

\n
\n

(1) There should be : len(train_df.user_id.unique()) == len(train_df[train_df.timestamp==0])

\n
\n

which comes from this assumption : (2) one user's row => one unique timestamp

\n

Am I right ?

\n

If so, I'm arguing that (2) isn't correct, because the data actually verify : one unique timestamp => one unique task_container_id => one or several rows

\n

Implying that (1) isn't correct too, as there could be timestamp==0 for several rows corresponding to a unique user. For example, the first interaction of a user could be : answering to 3 questions. That's why I mentioned task_container_id :)

\n

Tell me if we are on the same page !

", "rawMarkdown": "If I understand you well, you're assuming : \n> (1) There should be : len(train_df.user_id.unique()) == len(train_df[train_df.timestamp==0])\n\nwhich comes from this assumption : `(2) one user's row => one unique timestamp `\n\nAm I right ?\n\nIf so, I'm arguing that (2) isn't correct, because the data actually verify : `one unique timestamp => one unique task_container_id => one or several rows `\n\nImplying that (1) isn't correct too, as there could be timestamp==0 for several rows corresponding to a unique user. For example, the first interaction of a user could be : answering to 3 questions. That's why I mentioned task_container_id :) \n\nTell me if we are on the same page !", "votes": 1 }, { "id": 1100023, "postDate": "2020-12-02T19:13:15.477Z", "content": "
train_df.user_id.nunique() \n393656\n\ntrain_df.groupby('user_id').agg(ts=('timestamp','min'))-393656\ntrain_df.groupby('user_id').agg(ts=('timestamp','min')) .sum() - 0 \n
\n

So i think this sets assumption right,maybe some records earlier got filtered.

", "rawMarkdown": "```\ntrain_df.user_id.nunique() \n393656\n\ntrain_df.groupby('user_id').agg(ts=('timestamp','min'))-393656\ntrain_df.groupby('user_id').agg(ts=('timestamp','min')) .sum() - 0 \n\n```\nSo i think this sets assumption right,maybe some records earlier got filtered." }, { "id": 1132857, "postDate": "2020-12-30T18:21:55.030Z", "content": "
\n

As discussed here, timestamp is the time when a task (= unique task_container_id) is completed by a user. So train_df[train_df.timestamp==0] selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.
\n By the way, we don't always have task_container_id==0 when timestamp==0, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).

\n
\n

@johannhuber after long time.. i have one question.. do we know why do they set timestamp=0 instead of marking it as Elapsed time +0 as per definition of time stamp.

\n

However Time stamp definition is still confusing when comparing it with actual meaning

\n

It says Timestamp in milliseconds between this user interaction and time of completion of first even from user

\n

How does this definition maps to actual meaning.
\nthis user interaction -current question ? then which time period are we talking about here.

", "rawMarkdown": "> As discussed [here](https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189465), timestamp is the time when a task (= unique task_container_id) is completed by a user. So `train_df[train_df.timestamp==0]` selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.\n> By the way, we don't always have `task_container_id==0` when `timestamp==0`, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).\n\n@johannhuber after long time.. i have one question.. do we know why do they set timestamp=0 instead of marking it as Elapsed time +0 as per definition of time stamp. \n\nHowever Time stamp definition is still confusing when comparing it with actual meaning\n\n`It says Timestamp in milliseconds between this user interaction and time of completion of first even from user `\n\nHow does this definition maps to actual meaning.\nthis user interaction -current question ? then which time period are we talking about here." } ] }, { "id": 1133036, "postDate": "2020-12-30T21:40:39.200Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1099622, "author_name": "JT-karl", "author_url": "", "post_date": "2020-12-02T13:37:13.497000", "content": "

As discussed here, timestamp is the time when a task (= unique task_container_id) is completed by a user. So train_df[train_df.timestamp==0] selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.
\nBy the way, we don't always have task_container_id==0 when timestamp==0, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).

", "votes": 1, "replies": [ { "id": 1099690, "author_name": "Jaideep", "author_url": "", "post_date": "2020-12-02T14:35:54.653000", "content": "

@johannhuber i checked the discussion it talks on Task container id order more .
\nHere the doubt is why Not all User ids have Timestamp starting 0 , so in next task or interaction we see the time user took to complete the set of previous tasks..

", "votes": 0, "replies": [] }, { "id": 1099701, "author_name": "JT-karl", "author_url": "", "post_date": "2020-12-02T14:47:43.387000", "content": "

Then I'm not sure to understand your doubt : what do you mean by \"Not all User ids have Timestamp starting 0\" ?
\nActually I got :
\ntrain.groupby(['user_id'])['timestamp'].first().unique() # array([0])
\ntrain.groupby(['user_id'])['timestamp'].min().unique() # array([0])

", "votes": 1, "replies": [] }, { "id": 1099714, "author_name": "Jaideep", "author_url": "", "post_date": "2020-12-02T14:58:44.817000", "content": "

you can check my query above for DF.. number of unique users , number rows with Timestamp==0 should come as same, please see if you getting it same

", "votes": 0, "replies": [] }, { "id": 1099836, "author_name": "JT-karl", "author_url": "", "post_date": "2020-12-02T16:13:31.733000", "content": "

If I understand you well, you're assuming :

\n
\n

(1) There should be : len(train_df.user_id.unique()) == len(train_df[train_df.timestamp==0])

\n
\n

which comes from this assumption : (2) one user's row => one unique timestamp

\n

Am I right ?

\n

If so, I'm arguing that (2) isn't correct, because the data actually verify : one unique timestamp => one unique task_container_id => one or several rows

\n

Implying that (1) isn't correct too, as there could be timestamp==0 for several rows corresponding to a unique user. For example, the first interaction of a user could be : answering to 3 questions. That's why I mentioned task_container_id :)

\n

Tell me if we are on the same page !

", "votes": 1, "replies": [] }, { "id": 1100023, "author_name": "Jaideep", "author_url": "", "post_date": "2020-12-02T19:13:15.477000", "content": "
train_df.user_id.nunique() \n393656\n\ntrain_df.groupby('user_id').agg(ts=('timestamp','min'))-393656\ntrain_df.groupby('user_id').agg(ts=('timestamp','min')) .sum() - 0 \n
\n

So i think this sets assumption right,maybe some records earlier got filtered.

", "votes": 0, "replies": [] }, { "id": 1132857, "author_name": "Jaideep", "author_url": "", "post_date": "2020-12-30T18:21:55.030000", "content": "
\n

As discussed here, timestamp is the time when a task (= unique task_container_id) is completed by a user. So train_df[train_df.timestamp==0] selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.
\n By the way, we don't always have task_container_id==0 when timestamp==0, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).

\n
\n

@johannhuber after long time.. i have one question.. do we know why do they set timestamp=0 instead of marking it as Elapsed time +0 as per definition of time stamp.

\n

However Time stamp definition is still confusing when comparing it with actual meaning

\n

It says Timestamp in milliseconds between this user interaction and time of completion of first even from user

\n

How does this definition maps to actual meaning.
\nthis user interaction -current question ? then which time period are we talking about here.

", "votes": 0, "replies": [] } ] }, { "id": 1133036, "author_name": "", "author_url": "", "post_date": "2020-12-30T21:40:39.200000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1099473": "Timestamp =0 means if m correct ,time from which the user interaction clock starts . \n\nHowever there are some doubts i have\nEach user should have Timestamp staring at 0 . so number of Timestam=0 should be come as Number of unique users. \nBut i think that is not the case\n```\ntrain_df.user_id.nunique()\n123365\n```\n\n```\ntrain_df[train_df.timestamp==0]\n124586 rows × 9 columns\n```", "1099622": "As discussed [here](https://www.kaggle.com/c/riiid-test-answer-prediction/discussion/189465), timestamp is the time when a task (= unique task_container_id) is completed by a user. So `train_df[train_df.timestamp==0]` selects all rows corresponding to the first completed interactions - some of them are a bunch of questions from a single user.\nBy the way, we don't always have `task_container_id==0` when `timestamp==0`, because container ids are allocated when the user first sees a task (from @aquatic in the aboved linked thread).", "1133036": "" } }