{ "id": 503219, "title": "Some thoughts from an analysis of R2 per target", "url": "/competitions/leap-atmospheric-physics-ai-climsim/discussion/503219", "author_name": "Federico Peccia", "post_date": "2024-05-16T14:17:52.757000", "votes": 5, "comment_count": 4, "views": 0, "content": "

In my tests, if I analyze the R2 per target, for example for the ptend_t variables, I am able to get the following plot after training:

\n

\n

This looks very similar to the same curve they present on the paper:

\n

\n

It seems that the ptend_t targets around 20 and around 50 are not that straightforward to predict. I would be interested in knowing if you are able to get better R2 scores for these targets, or if your curve looks different than mine (specially if you are scoring better than the baseline ;) ).

\n

Thanks!

", "messages": [ { "id": 2816746, "postDate": "2024-05-16T14:17:52.757Z", "content": "

In my tests, if I analyze the R2 per target, for example for the ptend_t variables, I am able to get the following plot after training:

\n

\n

This looks very similar to the same curve they present on the paper:

\n

\n

It seems that the ptend_t targets around 20 and around 50 are not that straightforward to predict. I would be interested in knowing if you are able to get better R2 scores for these targets, or if your curve looks different than mine (specially if you are scoring better than the baseline ;) ).

\n

Thanks!

", "rawMarkdown": "In my tests, if I analyze the R2 per target, for example for the ptend_t variables, I am able to get the following plot after training:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1583716%2F60980007bf3b38b621a0f4e9eaba7801%2FScreenshot%20from%202024-05-16%2016-17-07.png?generation=1715869046410824&alt=media)\n\nThis looks very similar to the same curve they present on the paper:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1583716%2F104505275a0b384a8a4553e3f54846ba%2FScreenshot%20from%202024-05-16%2016-14-51.png?generation=1715868903063898&alt=media)\n\nIt seems that the ptend_t targets around 20 and around 50 are not that straightforward to predict. I would be interested in knowing if you are able to get better R2 scores for these targets, or if your curve looks different than mine (specially if you are scoring better than the baseline ;) ). \n\nThanks!", "votes": 5 }, { "id": 2889185, "postDate": "2024-06-25T10:06:15.337Z", "content": "

I have obtained the same structure, but there is a shift towards lower side in r2-graph.
\n(around ptend_q0002_17~25 are negative)
\nHave you also experienced such shift?

", "rawMarkdown": "I have obtained the same structure, but there is a shift towards lower side in r2-graph.\n(around ptend_q0002_17~25 are negative)\nHave you also experienced such shift?\n" }, { "id": 2817081, "postDate": "2024-05-16T17:28:31.493Z", "content": "

I have a similar curve.

\n

", "rawMarkdown": "I have a similar curve. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2F3535c518a54ab9023297c5207c081ea8%2F231564356234234.png?generation=1715880368457908&alt=media)", "replies": [ { "id": 2817351, "postDate": "2024-05-16T20:56:58.957Z", "content": "

Interesting! Thank you very much for sharing!

", "rawMarkdown": "Interesting! Thank you very much for sharing!" } ] }, { "id": 2890378, "postDate": "2024-06-26T05:04:36.317Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2889185, "author_name": "MBOOK", "author_url": "", "post_date": "2024-06-25T10:06:15.337000", "content": "

I have obtained the same structure, but there is a shift towards lower side in r2-graph.
\n(around ptend_q0002_17~25 are negative)
\nHave you also experienced such shift?

", "votes": 0, "replies": [] }, { "id": 2817081, "author_name": "Urazalinov Baurzhan", "author_url": "", "post_date": "2024-05-16T17:28:31.493000", "content": "

I have a similar curve.

\n

", "votes": 0, "replies": [ { "id": 2817351, "author_name": "Federico Peccia", "author_url": "", "post_date": "2024-05-16T20:56:58.957000", "content": "

Interesting! Thank you very much for sharing!

", "votes": 0, "replies": [] } ] }, { "id": 2890378, "author_name": "", "author_url": "", "post_date": "2024-06-26T05:04:36.317000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2816746": "In my tests, if I analyze the R2 per target, for example for the ptend_t variables, I am able to get the following plot after training:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1583716%2F60980007bf3b38b621a0f4e9eaba7801%2FScreenshot%20from%202024-05-16%2016-17-07.png?generation=1715869046410824&alt=media)\n\nThis looks very similar to the same curve they present on the paper:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1583716%2F104505275a0b384a8a4553e3f54846ba%2FScreenshot%20from%202024-05-16%2016-14-51.png?generation=1715868903063898&alt=media)\n\nIt seems that the ptend_t targets around 20 and around 50 are not that straightforward to predict. I would be interested in knowing if you are able to get better R2 scores for these targets, or if your curve looks different than mine (specially if you are scoring better than the baseline ;) ). \n\nThanks!", "2889185": "I have obtained the same structure, but there is a shift towards lower side in r2-graph.\n(around ptend_q0002_17~25 are negative)\nHave you also experienced such shift?\n", "2817081": "I have a similar curve. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8162991%2F3535c518a54ab9023297c5207c081ea8%2F231564356234234.png?generation=1715880368457908&alt=media)", "2890378": "" } }