Am I correct that only 21 classes listed in scored_birds.json will be scored? So other classes can be ignored?
", "messages": [ { "id": 1692876, "postDate": "2022-02-16T10:06:49.063Z", "content": "Am I correct that only 21 classes listed in scored_birds.json will be scored? So other classes can be ignored?
", "rawMarkdown": "Am I correct that only 21 classes listed in scored_birds.json will be scored? So other classes can be ignored?", "votes": 23 }, { "id": 1693705, "postDate": "2022-02-16T22:26:49.473Z", "content": "Hi Araik! Correct, only predictions for the 21 species listed in scored_birds.json are scored for this competition, predictions for other species will be dropped from your submission.
", "rawMarkdown": "Hi Araik! Correct, only predictions for the 21 species listed in scored_birds.json are scored for this competition, predictions for other species will be dropped from your submission. ", "votes": 8, "replies": [ { "id": 1693951, "postDate": "2022-02-17T04:39:30.577Z", "content": "@amandanavine
\nHow did you choose the other 131 species in the train_audio dir ?
\nIn the previous competition, there were 397 species.
\nDoes that mean there are only 21 speicies of scoring targets and 131 nonscoring targets in the test data?
We only chose species that are likely to occur on Hawaii and had at least one eligible training sample, hence the other 131 species. Feel free to use datasets from previous editions, my guess is that training a successful bird call classifier can't be done with only 21 training classes. But maybe it indeed is possible, and you won't need training samples for other species after all….
", "rawMarkdown": "We only chose species that are likely to occur on Hawaii and had at least one eligible training sample, hence the other 131 species. Feel free to use datasets from previous editions, my guess is that training a successful bird call classifier can't be done with only 21 training classes. But maybe it indeed is possible, and you won't need training samples for other species after all....", "votes": 11 }, { "id": 1694259, "postDate": "2022-02-17T09:51:10.550Z", "content": "I see, thanks for the reply!
", "rawMarkdown": "I see, thanks for the reply!" }, { "id": 1694314, "postDate": "2022-02-17T10:42:52.280Z", "content": "@stefankahl
\nThank you for your reply.
\nAll test data was recorded in Hawaii?
Yes, exactly.
", "rawMarkdown": "Yes, exactly.", "votes": 4 }, { "id": 1694506, "postDate": "2022-02-17T13:32:31.317Z", "content": "Thanks!
", "rawMarkdown": "Thanks! " } ] }, { "id": 1694494, "postDate": "2022-02-17T13:19:38.097Z", "content": "There's not much data for these 21 classes - just 1266 entries (based on primary_label).
", "rawMarkdown": "There's not much data for these 21 classes - just 1266 entries (based on primary_label).", "votes": 2, "replies": [ { "id": 1694731, "postDate": "2022-02-17T17:06:53.300Z", "content": "Yes, indeed! We've found that often the most important species to monitor have less training data, as they are in habitats further from humans. For large multispecies classifiers, one can improve performance on a common 'core' of species with lots of data to improve top-line eval statistics, while species with less data available languish. This competition is intended to help us figure out how to best handle the long tail cases.
", "rawMarkdown": "Yes, indeed! We've found that often the most important species to monitor have less training data, as they are in habitats further from humans. For large multispecies classifiers, one can improve performance on a common 'core' of species with lots of data to improve top-line eval statistics, while species with less data available languish. This competition is intended to help us figure out how to best handle the [long tail cases](https://en.wikipedia.org/wiki/Resplendent_quetzal#/media/File:Respledent_Quetzal,_Costa_Rica_2016.jpg).", "votes": 10 } ] }, { "id": 1742915, "postDate": "2022-04-02T11:54:14.837Z", "content": "I see that for example afrsil1 is scored in the train_metadata.csv but I dont see it in the scored_birds.json file. So which ones are scored and which one are not? What I'm missing?
\nThanks you :)
", "rawMarkdown": "I see that for example afrsil1 is scored in the train_metadata.csv but I dont see it in the scored_birds.json file. So which ones are scored and which one are not? What I'm missing?\n\nThanks you :)", "replies": [ { "id": 1773028, "postDate": "2022-04-30T19:49:47.573Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1773073, "postDate": "2022-04-30T21:13:47.310Z", "content": "yes I saw it. It was a bit confusing because it is my first competition. Thanks you.
", "rawMarkdown": "yes I saw it. It was a bit confusing because it is my first competition. Thanks you." } ] } ], "comments": [ { "id": 1693705, "author_name": "Amanda Navine", "author_url": "", "post_date": "2022-02-16T22:26:49.473000", "content": "Hi Araik! Correct, only predictions for the 21 species listed in scored_birds.json are scored for this competition, predictions for other species will be dropped from your submission.
", "votes": 8, "replies": [ { "id": 1693951, "author_name": "UEMU", "author_url": "", "post_date": "2022-02-17T04:39:30.577000", "content": "@amandanavine
\nHow did you choose the other 131 species in the train_audio dir ?
\nIn the previous competition, there were 397 species.
\nDoes that mean there are only 21 speicies of scoring targets and 131 nonscoring targets in the test data?
We only chose species that are likely to occur on Hawaii and had at least one eligible training sample, hence the other 131 species. Feel free to use datasets from previous editions, my guess is that training a successful bird call classifier can't be done with only 21 training classes. But maybe it indeed is possible, and you won't need training samples for other species after all….
", "votes": 11, "replies": [] }, { "id": 1694259, "author_name": "Araik Tamazian", "author_url": "", "post_date": "2022-02-17T09:51:10.550000", "content": "I see, thanks for the reply!
", "votes": 0, "replies": [] }, { "id": 1694314, "author_name": "UEMU", "author_url": "", "post_date": "2022-02-17T10:42:52.280000", "content": "@stefankahl
\nThank you for your reply.
\nAll test data was recorded in Hawaii?
Yes, exactly.
", "votes": 4, "replies": [] }, { "id": 1694506, "author_name": "UEMU", "author_url": "", "post_date": "2022-02-17T13:32:31.317000", "content": "Thanks!
", "votes": 0, "replies": [] } ] }, { "id": 1694494, "author_name": "Araik Tamazian", "author_url": "", "post_date": "2022-02-17T13:19:38.097000", "content": "There's not much data for these 21 classes - just 1266 entries (based on primary_label).
", "votes": 2, "replies": [ { "id": 1694731, "author_name": "Tom Denton", "author_url": "", "post_date": "2022-02-17T17:06:53.300000", "content": "Yes, indeed! We've found that often the most important species to monitor have less training data, as they are in habitats further from humans. For large multispecies classifiers, one can improve performance on a common 'core' of species with lots of data to improve top-line eval statistics, while species with less data available languish. This competition is intended to help us figure out how to best handle the long tail cases.
", "votes": 10, "replies": [] } ] }, { "id": 1742915, "author_name": "José Luis Córdoba Cabanillas", "author_url": "", "post_date": "2022-04-02T11:54:14.837000", "content": "I see that for example afrsil1 is scored in the train_metadata.csv but I dont see it in the scored_birds.json file. So which ones are scored and which one are not? What I'm missing?
\nThanks you :)
", "votes": 0, "replies": [ { "id": 1773028, "author_name": "", "author_url": "", "post_date": "2022-04-30T19:49:47.573000", "content": "", "votes": 1, "replies": [] }, { "id": 1773073, "author_name": "José Luis Córdoba Cabanillas", "author_url": "", "post_date": "2022-04-30T21:13:47.310000", "content": "yes I saw it. It was a bit confusing because it is my first competition. Thanks you.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1692876": "Am I correct that only 21 classes listed in scored_birds.json will be scored? So other classes can be ignored?", "1693705": "Hi Araik! Correct, only predictions for the 21 species listed in scored_birds.json are scored for this competition, predictions for other species will be dropped from your submission. ", "1694494": "There's not much data for these 21 classes - just 1266 entries (based on primary_label).", "1742915": "I see that for example afrsil1 is scored in the train_metadata.csv but I dont see it in the scored_birds.json file. So which ones are scored and which one are not? What I'm missing?\n\nThanks you :)" } }