\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4426199%2F84c4f0a9e6a2811da1a0ba41a849def2%2FScreen%20Shot%202020-10-26%20at%206.24.36%20PM.png?generation=1603733843142895&alt=media\")
I took a look at how average correctness per student evolves over time:
\n
And I found one particular student that seems to stand out from the crowd:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4426199%2Fb175d9e0d71841364ac8a3f1686c9657%2FScreen%20Shot%202020-10-26%20at%206.25.48%20PM.png?generation=1603733906843311&alt=media\")
It seems like the student went through the entire curriculum of learning tasks within 2-3 days after starting:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4426199%2F5ad686614279d0e785ded718c40cea1d%2FScreen%20Shot%202020-10-26%20at%206.26.35%20PM.png?generation=1603733924190052&alt=media\")
And the average correctness was around 0.27, indicating that the person kept randomly clicking one of the 4 multiple choice options without even reading the question.
\nWould it make sense to remove such students from the dataset before training the model?
\nThe student's ID is 7171715 for those interested.
I took a look at how average correctness per student evolves over time:
\n
And I found one particular student that seems to stand out from the crowd:
\n
It seems like the student went through the entire curriculum of learning tasks within 2-3 days after starting:
\n
And the average correctness was around 0.27, indicating that the person kept randomly clicking one of the 4 multiple choice options without even reading the question.
\nWould it make sense to remove such students from the dataset before training the model?
\nThe student's ID is 7171715 for those interested.
Beautiful plots! Which module is it by the way? Doesn't seem like plotly/bokeh to me. Ty for the info!
", "rawMarkdown": "Beautiful plots! Which module is it by the way? Doesn't seem like plotly/bokeh to me. Ty for the info!", "votes": null }, { "id": "1061077", "postDate": "10/26/2020 18:12:41", "content": "These were generated with Tableau.
", "rawMarkdown": "These were generated with Tableau.", "votes": null }, { "id": "1061143", "postDate": "10/26/2020 19:12:57", "content": "Why not to predict 0.27 all the time for him? 😄
", "rawMarkdown": "Why not to predict 0.27 all the time for him? 😄", "votes": null }, { "id": "1064126", "postDate": "10/29/2020 18:34:46", "content": "Some students just have too much free time 😄
", "rawMarkdown": "Some students just have too much free time 😄", "votes": null }, { "id": "1092178", "postDate": "11/26/2020 15:26:46", "content": "If the outliers are very few in number compared to the total data size, your models shouldn't be affected too much. Do be careful when there are multiple outliers!
", "rawMarkdown": "If the outliers are very few in number compared to the total data size, your models shouldn't be affected too much. Do be careful when there are multiple outliers!", "votes": null }, { "id": "1092210", "postDate": "11/26/2020 15:57:16", "content": "Interesting finding. What if the user_id 7171715 appears in the test set？ I think it should be remained.
", "rawMarkdown": "Interesting finding. What if the user_id 7171715 appears in the test set？ I think it should be remained.", "votes": null } ], "comments": [ { "id": 1061069, "author_name": "adityaecdrid", "author_url": "", "post_date": "10/26/2020 18:06:58", "content": "Beautiful plots! Which module is it by the way? Doesn't seem like plotly/bokeh to me. Ty for the info!
", "votes": null, "replies": [ { "id": 1061077, "author_name": "nyakaggle", "author_url": "", "post_date": "10/26/2020 18:12:41", "content": "These were generated with Tableau.
", "votes": null, "replies": [] }, { "id": 1061143, "author_name": "yihdarshieh", "author_url": "", "post_date": "10/26/2020 19:12:57", "content": "Why not to predict 0.27 all the time for him? 😄
", "votes": null, "replies": [] } ] }, { "id": 1064126, "author_name": "rafiko1", "author_url": "", "post_date": "10/29/2020 18:34:46", "content": "Some students just have too much free time 😄
", "votes": null, "replies": [] }, { "id": 1092178, "author_name": "namansood", "author_url": "", "post_date": "11/26/2020 15:26:46", "content": "If the outliers are very few in number compared to the total data size, your models shouldn't be affected too much. Do be careful when there are multiple outliers!
", "votes": null, "replies": [] }, { "id": 1092210, "author_name": "kazekage", "author_url": "", "post_date": "11/26/2020 15:57:16", "content": "Interesting finding. What if the user_id 7171715 appears in the test set？ I think it should be remained.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1061025": "I took a look at how average correctness per student evolves over time:\n\n\nAnd I found one particular student that seems to stand out from the crowd:\n\n\nIt seems like the student went through the _entire curriculum_ of learning tasks within 2-3 days after starting:\n\n\nAnd the average correctness was around 0.27, indicating that the person kept randomly clicking one of the 4 multiple choice options without even reading the question.\n\nWould it make sense to remove such students from the dataset before training the model?\n\nThe student's ID is `7171715` for those interested.", "1061069": "Beautiful plots! Which module is it by the way? Doesn't seem like plotly/bokeh to me. Ty for the info!", "1061077": "These were generated with Tableau.", "1061143": "Why not to predict 0.27 all the time for him? 😄", "1064126": "Some students just have too much free time 😄", "1092178": "If the outliers are very few in number compared to the total data size, your models shouldn't be affected too much. Do be careful when there are multiple outliers!", "1092210": "Interesting finding. What if the user_id 7171715 appears in the test set？ I think it should be remained." }, "source": "meta" }