{ "id": 503168, "title": "The Elephant in the room: Will someone publish a notebook with a method to retrieve WEEK_NUM? ", "url": "/competitions/home-credit-credit-risk-model-stability/discussion/503168", "author_name": "narsil (jobs-in-data.com)", "post_date": "2024-05-16T09:29:55.888000", "votes": 6, "comment_count": 9, "views": 0, "content": "

No… wait, you should do it 1 day before the competition ends.

", "messages": [ { "id": 2816368, "postDate": "2024-05-16T09:29:55.887Z", "content": "

No… wait, you should do it 1 day before the competition ends.

", "rawMarkdown": "No... wait, you should do it 1 day before the competition ends.", "votes": 6 }, { "id": 2816436, "postDate": "2024-05-16T10:11:30.793Z", "content": "

I assume most willl just keep it to themselves to secure a medal spot easily..

\n

I don't think it makes sense for hosts to allow retrieving week_num if you can, but then hide it from most participants, so now for the last 2 weeks people will not focus on model improvements but rather reversing the week_num..

\n

If I reverse it sure I will publish it to level playing field for everyone, but not so motivated to try, next time I will read the comments before joining a competition so late 😅

", "rawMarkdown": "I assume most willl just keep it to themselves to secure a medal spot easily.. \n\nI don't think it makes sense for hosts to allow retrieving week_num if you can, but then hide it from most participants, so now for the last 2 weeks people will not focus on model improvements but rather reversing the week_num.. \n\nIf I reverse it sure I will publish it to level playing field for everyone, but not so motivated to try, next time I will read the comments before joining a competition so late 😅", "votes": 1 }, { "id": 2818366, "postDate": "2024-05-17T13:21:31.800Z", "content": "

The solution shared in the Code section is based on adversarial validation, which can identify the first half and second half of the test set, but cannot recover the WEEK_NUM.

", "rawMarkdown": "The [solution ](https://www.kaggle.com/code/tritionalval/this-is-the-way/notebook) shared in the Code section is based on adversarial validation, which can identify the first half and second half of the test set, but cannot recover the WEEK_NUM.", "votes": 2, "replies": [ { "id": 2818423, "postDate": "2024-05-17T14:02:24.423Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2819617, "postDate": "2024-05-17T15:50:25.363Z", "content": "

I tested this kind of solution as well. But If you are overfitting the leaderbord your going to do it heavyly. So it is kind of risky just to rely on that.

", "rawMarkdown": "I tested this kind of solution as well. But If you are overfitting the leaderbord your going to do it heavyly. So it is kind of risky just to rely on that.", "votes": 1 }, { "id": 2821217, "postDate": "2024-05-18T02:30:34.123Z", "content": "

Why do you claim that this method can identify the first half and second half of the test set?

\n
condition=y_pred<\nf_subm.loc[condition, ] = (df_subm.loc[condition, ] - ).clip()\n
\n

Isn't it just coincidently that this code worsens the first-half score in public LB?

", "rawMarkdown": "Why do you claim that this method can identify the first half and second half of the test set?\n```python\ncondition=y_pred<0.96\nf_subm.loc[condition, 'score'] = (df_subm.loc[condition, 'score'] - 0.05).clip(0)\n```\nIsn't it just coincidently that this code worsens the first-half score in public LB?", "isDeleted": true } ] }, { "id": 2821599, "postDate": "2024-05-18T07:03:16.463Z", "content": "

A description how to do it, would be enough 😃

", "rawMarkdown": "A description how to do it, would be enough 😃", "votes": -3 }, { "id": 2816804, "postDate": "2024-05-16T14:40:38.640Z", "content": "

I think in general people publish notebooks to get upvotes. It's possible that if one publishes a notebook too late for people to notice before the competition ends, the interest will almost entirely disappear after the deadline. So why bother do it at all on the last day?

", "rawMarkdown": "I think in general people publish notebooks to get upvotes. It's possible that if one publishes a notebook too late for people to notice before the competition ends, the interest will almost entirely disappear after the deadline. So why bother do it at all on the last day? " }, { "id": 2816582, "postDate": "2024-05-16T12:08:49.640Z", "content": "

A description how to do it, would be enough 😃

", "rawMarkdown": "A description how to do it, would be enough 😃" }, { "id": 2816443, "postDate": "2024-05-16T10:17:39.167Z", "rawMarkdown": "", "votes": -3, "isDeleted": true } ], "comments": [ { "id": 2816436, "author_name": "JM", "author_url": "", "post_date": "2024-05-16T10:11:30.793000", "content": "

I assume most willl just keep it to themselves to secure a medal spot easily..

\n

I don't think it makes sense for hosts to allow retrieving week_num if you can, but then hide it from most participants, so now for the last 2 weeks people will not focus on model improvements but rather reversing the week_num..

\n

If I reverse it sure I will publish it to level playing field for everyone, but not so motivated to try, next time I will read the comments before joining a competition so late 😅

", "votes": 1, "replies": [] }, { "id": 2818366, "author_name": "Bruce", "author_url": "", "post_date": "2024-05-17T13:21:31.800000", "content": "

The solution shared in the Code section is based on adversarial validation, which can identify the first half and second half of the test set, but cannot recover the WEEK_NUM.

", "votes": 2, "replies": [ { "id": 2818423, "author_name": "", "author_url": "", "post_date": "2024-05-17T14:02:24.423000", "content": "", "votes": 0, "replies": [] }, { "id": 2819617, "author_name": "LucaMTB", "author_url": "", "post_date": "2024-05-17T15:50:25.363000", "content": "

I tested this kind of solution as well. But If you are overfitting the leaderbord your going to do it heavyly. So it is kind of risky just to rely on that.

", "votes": 1, "replies": [] }, { "id": 2821217, "author_name": "", "author_url": "", "post_date": "2024-05-18T02:30:34.123000", "content": "

Why do you claim that this method can identify the first half and second half of the test set?

\n
condition=y_pred<\nf_subm.loc[condition, ] = (df_subm.loc[condition, ] - ).clip()\n
\n

Isn't it just coincidently that this code worsens the first-half score in public LB?

", "votes": 0, "replies": [] } ] }, { "id": 2821599, "author_name": "Harsh Upadhyay", "author_url": "", "post_date": "2024-05-18T07:03:16.463000", "content": "

A description how to do it, would be enough 😃

", "votes": -3, "replies": [] }, { "id": 2816804, "author_name": "Eduard Stefanescu", "author_url": "", "post_date": "2024-05-16T14:40:38.640000", "content": "

I think in general people publish notebooks to get upvotes. It's possible that if one publishes a notebook too late for people to notice before the competition ends, the interest will almost entirely disappear after the deadline. So why bother do it at all on the last day?

", "votes": 0, "replies": [] }, { "id": 2816582, "author_name": "LucaMTB", "author_url": "", "post_date": "2024-05-16T12:08:49.640000", "content": "

A description how to do it, would be enough 😃

", "votes": 0, "replies": [] }, { "id": 2816443, "author_name": "", "author_url": "", "post_date": "2024-05-16T10:17:39.167000", "content": "", "votes": -3, "replies": [] } ], "raw_markdown_by_id": { "2816368": "No... wait, you should do it 1 day before the competition ends.", "2816436": "I assume most willl just keep it to themselves to secure a medal spot easily.. \n\nI don't think it makes sense for hosts to allow retrieving week_num if you can, but then hide it from most participants, so now for the last 2 weeks people will not focus on model improvements but rather reversing the week_num.. \n\nIf I reverse it sure I will publish it to level playing field for everyone, but not so motivated to try, next time I will read the comments before joining a competition so late 😅", "2818366": "The [solution ](https://www.kaggle.com/code/tritionalval/this-is-the-way/notebook) shared in the Code section is based on adversarial validation, which can identify the first half and second half of the test set, but cannot recover the WEEK_NUM.", "2821599": "A description how to do it, would be enough 😃", "2816804": "I think in general people publish notebooks to get upvotes. It's possible that if one publishes a notebook too late for people to notice before the competition ends, the interest will almost entirely disappear after the deadline. So why bother do it at all on the last day? ", "2816582": "A description how to do it, would be enough 😃", "2816443": "" } }