\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12870466%2Fe9e785406498dc9a6546a7ea34b889fd%2FSession%20Id.drawio.png?generation=1680446585768163&alt=media\")
\nI made an extended EDA on this specific point here, here is a something that I haven't seen been exploited at all in the models:
\n
As we now know thx to @pdnartreb, we can get the session start date and time of the day:
\n
\nI made an extended EDA on this specific point here, here is a something that I haven't seen been exploited at all in the models:
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F12870466%2F9aebbab61d3e7a5a1e58855476c7eb76%2FScreenshot_79.jpg?generation=1680448453575348&alt=media\")
I think the session data has been underused and that there is a lot that can be found using that information..
", "messages": [ { "id": "2195974", "postDate": "03/25/2023 02:53:19", "content": "As we now know thx to @pdnartreb, we can get the session start date and time of the day:
\n
\nI made an extended EDA on this specific point here, here is a something that I haven't seen been exploited at all in the models:
\n
I think the session data has been underused and that there is a lot that can be found using that information..
", "rawMarkdown": "As we now know thx to @pdnartreb, we can get the session start date and time of the day:\n\nI made an **extended EDA on this specific point [here](https://www.kaggle.com/code/janmpia/session-start-time-eda)**, here is a something that I haven't seen been exploited at all in the models: \n\n\n\n\nI think the session data has been underused and that there is a lot that can be found using that information..", "votes": null }, { "id": "2196138", "postDate": "03/25/2023 06:48:01", "content": "Well, that's an interesting discovery! I am wondering why the acc is so low in 2020-10, perhaps the game was just released and players are trying to get familiarized with it?
", "rawMarkdown": "Well, that's an interesting discovery! I am wondering why the acc is so low in 2020-10, perhaps the game was just released and players are trying to get familiarized with it?", "votes": null }, { "id": "2196646", "postDate": "03/25/2023 14:59:19", "content": "I think that is because of the very low number of values for that time period, if you check the notebook, you'll see a lot of extreme values for this specifique date !
", "rawMarkdown": "I think that is because of the very low number of values for that time period, if you check the notebook, you'll see a lot of extreme values for this specifique date !", "votes": null }, { "id": "2197864", "postDate": "03/26/2023 13:54:26", "content": "yeah, you are right, perhaps the game was not so popular at that time.
", "rawMarkdown": "yeah, you are right, perhaps the game was not so popular at that time.", "votes": null }, { "id": "2197902", "postDate": "03/26/2023 14:36:07", "content": "or they released it later that month, since for the second month it was way more instantly.
", "rawMarkdown": "or they released it later that month, since for the second month it was way more instantly.", "votes": null }, { "id": "2198240", "postDate": "03/26/2023 19:33:40", "content": "This is indeed interesting
", "rawMarkdown": "This is indeed interesting", "votes": null }, { "id": "2201074", "postDate": "03/29/2023 03:01:09", "content": "Have you been able to extract anything from it ? From what I have tested, it is only interesting for 2 questions, the rest is just pure noise I guess.
", "rawMarkdown": "Have you been able to extract anything from it ? From what I have tested, it is only interesting for 2 questions, the rest is just pure noise I guess.", "votes": null }, { "id": "2201883", "postDate": "03/29/2023 15:52:30", "content": "I think the time of day stuff in your notebook makes lot of sense (by the way, I really like how you used the session counts to confirm that the session_id contains date and time information), though as you show the differences are pretty small between class hours and out-of-class hours. The fact there is a significant difference in terms of length of sessions seems interesting though, and might suggest it's worth exploring further. Have you managed to dig much more up?
\nWith the above plots, I'm not sure I follow what might be driving the variance outside of noise, as you suggest. I assume the two questions you refer to are question five and six that you've also plotted over time? If there is something meaningful going on with those questions, what do you think might be causing it?
", "rawMarkdown": "I think the time of day stuff in your notebook makes lot of sense (by the way, I really like how you used the session counts to confirm that the session_id contains date and time information), though as you show the differences are pretty small between class hours and out-of-class hours. The fact there is a significant difference in terms of length of sessions seems interesting though, and might suggest it's worth exploring further. Have you managed to dig much more up? \n\nWith the above plots, I'm not sure I follow what might be driving the variance outside of noise, as you suggest. I assume the two questions you refer to are question five and six that you've also plotted over time? If there is something meaningful going on with those questions, what do you think might be causing it?", "votes": null }, { "id": "2203060", "postDate": "03/30/2023 13:38:23", "content": "Yes I was referring to the ones I showed on the plots, I changed the threshold a bit for those two predictions since I saw an upward trend to test and it turned out to work. I'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.
", "rawMarkdown": "Yes I was referring to the ones I showed on the plots, I changed the threshold a bit for those two predictions since I saw an upward trend to test and it turned out to work. I'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.", "votes": null }, { "id": "2203123", "postDate": "03/30/2023 14:22:20", "content": "\n\nI'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.
\n
Yeah I think you might be right. Only possibility I can think of is whether or not there might be a small number of repeat users and those questions might be particularly memorable (can't remember what questions five and six were)?
", "rawMarkdown": ">I'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.\n\nYeah I think you might be right. Only possibility I can think of is whether or not there might be a small number of repeat users and those questions might be particularly memorable (can't remember what questions five and six were)?", "votes": null }, { "id": "2203332", "postDate": "03/30/2023 18:05:39", "content": "Neither can I, I'll check tomorrow first hour and update you here
", "rawMarkdown": "Neither can I, I'll check tomorrow first hour and update you here", "votes": null }, { "id": "2205553", "postDate": "04/01/2023 17:01:17", "content": "@paulrjohnson
\nQuestion 5:
\n\nQ_5 : \"Who told me it was a dry cleaning slip?\"
\n
\n A_5 : Clothes Expert [Notebook fourth page, second item]
Question 6:
\n\n\nQ_6 : \"How did I find the right dry cleaner?\"
\n
\n A_6 : Squeaky Cleaners card [Notebook fifth page, first item]
I have no idea why thoses ones in particular looking solely at the question,
\nif an idea comes to you I'll be happy to ear it
Haha. Yeah I have a hard time thinking of any reason why either of those questions might be notable enough that you'd see improved performance over time.
\nI think it's probably just noise!
", "rawMarkdown": "Haha. Yeah I have a hard time thinking of any reason why either of those questions might be notable enough that you'd see improved performance over time.\n\nI think it's probably just noise!", "votes": null }, { "id": "2206571", "postDate": "04/02/2023 16:55:44", "content": "changing the threshold does help the model tho, did you give it a try ?
", "rawMarkdown": "changing the threshold does help the model tho, did you give it a try ?", "votes": null }, { "id": "2209738", "postDate": "04/04/2023 21:20:10", "content": "Sorry, not sure I follow. What do you mean when you say changing the threshold?
", "rawMarkdown": "Sorry, not sure I follow. What do you mean when you say changing the threshold?", "votes": null }, { "id": "2223513", "postDate": "04/16/2023 10:45:48", "content": "Game is much older than 2020 - its github code dates back to 2017.
\nI suspect they started recording user data in 2020-10.
", "rawMarkdown": "Game is much older than 2020 - its github [code](https://github.com/fielddaylab/jo_wilder) dates back to 2017.\n\nI suspect they started recording user data in 2020-10.", "votes": null }, { "id": "2259337", "postDate": "05/14/2023 21:49:01", "content": "Thank you!
\ndate in session not equal to start session date:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4172517%2Fca44fa16af26d2eef7fef5c374b0adc2%2Fsome.png?generation=1684100913234204&alt=media\")
Well, that's an interesting discovery! I am wondering why the acc is so low in 2020-10, perhaps the game was just released and players are trying to get familiarized with it?
", "votes": null, "replies": [ { "id": 2196646, "author_name": "janmpia", "author_url": "", "post_date": "03/25/2023 14:59:19", "content": "I think that is because of the very low number of values for that time period, if you check the notebook, you'll see a lot of extreme values for this specifique date !
", "votes": null, "replies": [ { "id": 2197864, "author_name": "harrisonliang", "author_url": "", "post_date": "03/26/2023 13:54:26", "content": "yeah, you are right, perhaps the game was not so popular at that time.
", "votes": null, "replies": [ { "id": 2197902, "author_name": "janmpia", "author_url": "", "post_date": "03/26/2023 14:36:07", "content": "or they released it later that month, since for the second month it was way more instantly.
", "votes": null, "replies": [] } ] } ] }, { "id": 2223513, "author_name": "ngocuong", "author_url": "", "post_date": "04/16/2023 10:45:48", "content": "Game is much older than 2020 - its github code dates back to 2017.
\nI suspect they started recording user data in 2020-10.
", "votes": null, "replies": [] } ] }, { "id": 2198240, "author_name": "simonveitner", "author_url": "", "post_date": "03/26/2023 19:33:40", "content": "This is indeed interesting
", "votes": null, "replies": [ { "id": 2201074, "author_name": "janmpia", "author_url": "", "post_date": "03/29/2023 03:01:09", "content": "Have you been able to extract anything from it ? From what I have tested, it is only interesting for 2 questions, the rest is just pure noise I guess.
", "votes": null, "replies": [ { "id": 2201883, "author_name": "paulrjohnson", "author_url": "", "post_date": "03/29/2023 15:52:30", "content": "I think the time of day stuff in your notebook makes lot of sense (by the way, I really like how you used the session counts to confirm that the session_id contains date and time information), though as you show the differences are pretty small between class hours and out-of-class hours. The fact there is a significant difference in terms of length of sessions seems interesting though, and might suggest it's worth exploring further. Have you managed to dig much more up?
\nWith the above plots, I'm not sure I follow what might be driving the variance outside of noise, as you suggest. I assume the two questions you refer to are question five and six that you've also plotted over time? If there is something meaningful going on with those questions, what do you think might be causing it?
", "votes": null, "replies": [ { "id": 2203060, "author_name": "janmpia", "author_url": "", "post_date": "03/30/2023 13:38:23", "content": "Yes I was referring to the ones I showed on the plots, I changed the threshold a bit for those two predictions since I saw an upward trend to test and it turned out to work. I'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.
", "votes": null, "replies": [ { "id": 2203123, "author_name": "paulrjohnson", "author_url": "", "post_date": "03/30/2023 14:22:20", "content": "\n\nI'm pretty sure this is very likely to be overfitting or a coincidence. It needs deeper analysis to confirm this actually might be viable.
\n
Yeah I think you might be right. Only possibility I can think of is whether or not there might be a small number of repeat users and those questions might be particularly memorable (can't remember what questions five and six were)?
", "votes": null, "replies": [ { "id": 2203332, "author_name": "janmpia", "author_url": "", "post_date": "03/30/2023 18:05:39", "content": "Neither can I, I'll check tomorrow first hour and update you here
", "votes": null, "replies": [] }, { "id": 2205553, "author_name": "janmpia", "author_url": "", "post_date": "04/01/2023 17:01:17", "content": "@paulrjohnson
\nQuestion 5:
\n\nQ_5 : \"Who told me it was a dry cleaning slip?\"
\n
\n A_5 : Clothes Expert [Notebook fourth page, second item]
Question 6:
\n\n\nQ_6 : \"How did I find the right dry cleaner?\"
\n
\n A_6 : Squeaky Cleaners card [Notebook fifth page, first item]
I have no idea why thoses ones in particular looking solely at the question,
\nif an idea comes to you I'll be happy to ear it
Haha. Yeah I have a hard time thinking of any reason why either of those questions might be notable enough that you'd see improved performance over time.
\nI think it's probably just noise!
", "votes": null, "replies": [ { "id": 2206571, "author_name": "janmpia", "author_url": "", "post_date": "04/02/2023 16:55:44", "content": "changing the threshold does help the model tho, did you give it a try ?
", "votes": null, "replies": [ { "id": 2209738, "author_name": "paulrjohnson", "author_url": "", "post_date": "04/04/2023 21:20:10", "content": "Sorry, not sure I follow. What do you mean when you say changing the threshold?
", "votes": null, "replies": [] } ] } ] } ] } ] } ] } ] } ] } ] }, { "id": 2259337, "author_name": "", "author_url": "", "post_date": "05/14/2023 21:49:01", "content": "Thank you!
\ndate in session not equal to start session date:
\n