{ "id": 394227, "title": "What's the difference between Bird 2023 with the past series?", "url": "/competitions/birdclef-2023/discussion/394227", "author_name": "", "post_date": "2023-03-12T16:54:43.693095500Z", "votes": 21, "comment_count": 5, "views": 0, "content": "

I'm new for Bird Series Comps, can someone briefly explain the difference between Bird2023 and the past comps?

\n

Many thanks.

", "messages": [ { "id": "2178723", "postDate": "03/12/2023 16:54:43", "content": "

I'm new for Bird Series Comps, can someone briefly explain the difference between Bird2023 and the past comps?

\n

Many thanks.

", "rawMarkdown": "I'm new for Bird Series Comps, can someone briefly explain the difference between Bird2023 and the past comps?\n\nMany thanks.", "votes": null }, { "id": "2178728", "postDate": "03/12/2023 17:00:03", "content": "

Hello!

\n

The biggest changes from competitor perspective are:

\n

a) We have switched to CMAP metric, which is threshold free. Many of the most important optimizations in previous years were for better threshold choice; the new metric should focus the competition more on core model quality.

\n

b) There is a pretty strict time limit on model execution. This will limit the use of giant ensembles.

\n

c) Every year, we change up the target dataset. This year we're working with birds of East Africa, many of which have very little training data available.

\n

(oh, and (d), the prize is much larger. I suppose that might be of interest, too. :) )

\n

Happy hacking!

", "rawMarkdown": "Hello!\n\nThe biggest changes from competitor perspective are:\n\na) We have switched to CMAP metric, which is threshold free. Many of the most important optimizations in previous years were for better threshold choice; the new metric should focus the competition more on core model quality.\n\nb) There is a pretty strict time limit on model execution. This will limit the use of giant ensembles.\n\nc) Every year, we change up the target dataset. This year we're working with birds of East Africa, many of which have very little training data available.\n\n(oh, and (d), the prize is much larger. I suppose that might be of interest, too. :) )\n\nHappy hacking!", "votes": null }, { "id": "2179258", "postDate": "03/13/2023 04:09:48", "content": "

Thanks @tomdenton for the information

", "rawMarkdown": "Thanks @tomdenton for the information", "votes": null }, { "id": "2179265", "postDate": "03/13/2023 04:10:46", "content": "

Thanks @evilpsycho42 For raising a important discussion

", "rawMarkdown": "Thanks @evilpsycho42 For raising a important discussion", "votes": null }, { "id": "2179281", "postDate": "03/13/2023 04:15:11", "content": "

Very useful summary!! Thanks a lot , so this allows us to focus on the powerfull single model. 🎉🎉

", "rawMarkdown": "Very useful summary!! Thanks a lot , so this allows us to focus on the powerfull single model. 🎉🎉", "votes": null }, { "id": "2186159", "postDate": "03/17/2023 15:59:02", "content": "

Yes, so we can focus on the core of this competition.

", "rawMarkdown": "Yes, so we can focus on the core of this competition.", "votes": null } ], "comments": [ { "id": 2178728, "author_name": "tomdenton", "author_url": "", "post_date": "03/12/2023 17:00:03", "content": "

Hello!

\n

The biggest changes from competitor perspective are:

\n

a) We have switched to CMAP metric, which is threshold free. Many of the most important optimizations in previous years were for better threshold choice; the new metric should focus the competition more on core model quality.

\n

b) There is a pretty strict time limit on model execution. This will limit the use of giant ensembles.

\n

c) Every year, we change up the target dataset. This year we're working with birds of East Africa, many of which have very little training data available.

\n

(oh, and (d), the prize is much larger. I suppose that might be of interest, too. :) )

\n

Happy hacking!

", "votes": null, "replies": [ { "id": 2179258, "author_name": "shaikjasheen", "author_url": "", "post_date": "03/13/2023 04:09:48", "content": "

Thanks @tomdenton for the information

", "votes": null, "replies": [] }, { "id": 2179281, "author_name": "evilpsycho42", "author_url": "", "post_date": "03/13/2023 04:15:11", "content": "

Very useful summary!! Thanks a lot , so this allows us to focus on the powerfull single model. 🎉🎉

", "votes": null, "replies": [] } ] }, { "id": 2179265, "author_name": "shaikjasheen", "author_url": "", "post_date": "03/13/2023 04:10:46", "content": "

Thanks @evilpsycho42 For raising a important discussion

", "votes": null, "replies": [ { "id": 2186159, "author_name": "evilpsycho42", "author_url": "", "post_date": "03/17/2023 15:59:02", "content": "

Yes, so we can focus on the core of this competition.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2178723": "I'm new for Bird Series Comps, can someone briefly explain the difference between Bird2023 and the past comps?\n\nMany thanks.", "2178728": "Hello!\n\nThe biggest changes from competitor perspective are:\n\na) We have switched to CMAP metric, which is threshold free. Many of the most important optimizations in previous years were for better threshold choice; the new metric should focus the competition more on core model quality.\n\nb) There is a pretty strict time limit on model execution. This will limit the use of giant ensembles.\n\nc) Every year, we change up the target dataset. This year we're working with birds of East Africa, many of which have very little training data available.\n\n(oh, and (d), the prize is much larger. I suppose that might be of interest, too. :) )\n\nHappy hacking!", "2179258": "Thanks @tomdenton for the information", "2179265": "Thanks @evilpsycho42 For raising a important discussion", "2179281": "Very useful summary!! Thanks a lot , so this allows us to focus on the powerfull single model. 🎉🎉", "2186159": "Yes, so we can focus on the core of this competition." }, "source": "meta" }