{ "id": 491532, "title": "basic queries", "url": "/competitions/birdclef-2024/discussion/491532", "author_name": "", "post_date": "2024-04-06T07:39:11.720796900Z", "votes": 3, "comment_count": 7, "views": 0, "content": "

hello, I wanted to get started with this, but I have few doubts regarding the competition.
\nIn the submission file we have

\n

row_id: A slug of soundscape_[soundscape_id]_[end_time] for the prediction.
\n[bird_id]: There are 182 bird ID columns. You will need to predict the probability of the presence of each bird for each row.

\n

so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?

\n

and also, are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?

", "messages": [ { "id": "2738223", "postDate": "04/06/2024 07:39:11", "content": "

hello, I wanted to get started with this, but I have few doubts regarding the competition.
\nIn the submission file we have

\n

row_id: A slug of soundscape_[soundscape_id]_[end_time] for the prediction.
\n[bird_id]: There are 182 bird ID columns. You will need to predict the probability of the presence of each bird for each row.

\n

so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?

\n

and also, are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?

", "rawMarkdown": "hello, I wanted to get started with this, but I have few doubts regarding the competition.\nIn the submission file we have \n\nrow_id: A slug of soundscape_[soundscape_id]_[end_time] for the prediction.\n[bird_id]: There are 182 bird ID columns. You will need to predict the probability of the presence of each bird for each row.\n\nso are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?\n\nand also, are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?", "votes": null }, { "id": "2738456", "postDate": "04/06/2024 10:53:15", "content": "
\n

so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?

\n
\n

Yes, the [end_time] is 5, 10, … , 240. To be precise, there are 48 rows per audio file, each row having a unique row_id and 182 different bird classification results.

\n
\n

are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?

\n
\n

My understanding is that you only need to predict the presence of bird sounds. In other words, no classification is needed as to what type of voice it is. More detailed metadata (\"type\", \"latitude\", etc.) should only be considered if you can make good use of them.

", "rawMarkdown": "> so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?\n\nYes, the [end_time] is 5, 10, ... , 240. To be precise, there are 48 rows per audio file, each row having a unique row_id and 182 different bird classification results.\n\n> are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?\n\nMy understanding is that you only need to predict the presence of bird sounds. In other words, no classification is needed as to what type of voice it is. More detailed metadata (\"type\", \"latitude\", etc.) should only be considered if you can make good use of them.", "votes": null }, { "id": "2738509", "postDate": "04/06/2024 11:24:58", "content": "

Thx,
\nI was specially confused with the description \"Your challenge in this competition is to identify which birds are calling in recordings\", I will just ignore it as of now

", "rawMarkdown": "Thx, \nI was specially confused with the description \"Your challenge in this competition is to identify which birds are calling in recordings\", I will just ignore it as of now", "votes": null }, { "id": "2738895", "postDate": "04/06/2024 17:28:32", "content": "

Do we know if there is a bird call present in each 5-sec segment of the test soundscapes?

", "rawMarkdown": "Do we know if there is a bird call present in each 5-sec segment of the test soundscapes?", "votes": null }, { "id": "2739269", "postDate": "04/06/2024 23:13:35", "content": "

No, we do not know at what time the birds are singing, so we need to make predictions for all 5sec segments.

", "rawMarkdown": "No, we do not know at what time the birds are singing, so we need to make predictions for all 5sec segments.", "votes": null }, { "id": "2739286", "postDate": "04/06/2024 23:43:15", "content": "

Hmm, so I am assuming the 5-sec segments with no bird calls are ignored during evaluation?

", "rawMarkdown": "Hmm, so I am assuming the 5-sec segments with no bird calls are ignored during evaluation?", "votes": null }, { "id": "2739293", "postDate": "04/06/2024 23:53:40", "content": "

They will not be ignored. Since the evaluation metric is AUROC, submitting a high confidence at segments without calls will result in a worse score.

", "rawMarkdown": "They will not be ignored. Since the evaluation metric is AUROC, submitting a high confidence at segments without calls will result in a worse score.", "votes": null }, { "id": "2740143", "postDate": "04/07/2024 15:57:25", "content": "

All segments are scored. It's good to keep in mind that vocalizations can overlap; we consider each species output as an independent binary classification problem, as each window may have no birds, one bird, or many birds.

", "rawMarkdown": "All segments are scored. It's good to keep in mind that vocalizations can overlap; we consider each species output as an independent binary classification problem, as each window may have no birds, one bird, or many birds.", "votes": null } ], "comments": [ { "id": 2738456, "author_name": "ryotayoshinobu", "author_url": "", "post_date": "04/06/2024 10:53:15", "content": "
\n

so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?

\n
\n

Yes, the [end_time] is 5, 10, … , 240. To be precise, there are 48 rows per audio file, each row having a unique row_id and 182 different bird classification results.

\n
\n

are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?

\n
\n

My understanding is that you only need to predict the presence of bird sounds. In other words, no classification is needed as to what type of voice it is. More detailed metadata (\"type\", \"latitude\", etc.) should only be considered if you can make good use of them.

", "votes": null, "replies": [ { "id": 2738509, "author_name": "arunsensei", "author_url": "", "post_date": "04/06/2024 11:24:58", "content": "

Thx,
\nI was specially confused with the description \"Your challenge in this competition is to identify which birds are calling in recordings\", I will just ignore it as of now

", "votes": null, "replies": [] }, { "id": 2738895, "author_name": "brendanartley", "author_url": "", "post_date": "04/06/2024 17:28:32", "content": "

Do we know if there is a bird call present in each 5-sec segment of the test soundscapes?

", "votes": null, "replies": [ { "id": 2739269, "author_name": "ryotayoshinobu", "author_url": "", "post_date": "04/06/2024 23:13:35", "content": "

No, we do not know at what time the birds are singing, so we need to make predictions for all 5sec segments.

", "votes": null, "replies": [ { "id": 2739286, "author_name": "brendanartley", "author_url": "", "post_date": "04/06/2024 23:43:15", "content": "

Hmm, so I am assuming the 5-sec segments with no bird calls are ignored during evaluation?

", "votes": null, "replies": [ { "id": 2739293, "author_name": "ryotayoshinobu", "author_url": "", "post_date": "04/06/2024 23:53:40", "content": "

They will not be ignored. Since the evaluation metric is AUROC, submitting a high confidence at segments without calls will result in a worse score.

", "votes": null, "replies": [] }, { "id": 2740143, "author_name": "tomdenton", "author_url": "", "post_date": "04/07/2024 15:57:25", "content": "

All segments are scored. It's good to keep in mind that vocalizations can overlap; we consider each species output as an independent binary classification problem, as each window may have no birds, one bird, or many birds.

", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "2738223": "hello, I wanted to get started with this, but I have few doubts regarding the competition.\nIn the submission file we have \n\nrow_id: A slug of soundscape_[soundscape_id]_[end_time] for the prediction.\n[bird_id]: There are 182 bird ID columns. You will need to predict the probability of the presence of each bird for each row.\n\nso are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?\n\nand also, are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?", "2738456": "> so are the predictions supposed to be on 5 sec intervals from the 4 min audio file ? if so then for a given row ID we have 48 different predictions, right?\n\nYes, the [end_time] is 5, 10, ... , 240. To be precise, there are 48 rows per audio file, each row having a unique row_id and 182 different bird classification results.\n\n> are we supposed to identify the bird calling and not just their presence? so does that mean in the train_matadata file type ['call'] has more significance?\n\nMy understanding is that you only need to predict the presence of bird sounds. In other words, no classification is needed as to what type of voice it is. More detailed metadata (\"type\", \"latitude\", etc.) should only be considered if you can make good use of them.", "2738509": "Thx, \nI was specially confused with the description \"Your challenge in this competition is to identify which birds are calling in recordings\", I will just ignore it as of now", "2738895": "Do we know if there is a bird call present in each 5-sec segment of the test soundscapes?", "2739269": "No, we do not know at what time the birds are singing, so we need to make predictions for all 5sec segments.", "2739286": "Hmm, so I am assuming the 5-sec segments with no bird calls are ignored during evaluation?", "2739293": "They will not be ignored. Since the evaluation metric is AUROC, submitting a high confidence at segments without calls will result in a worse score.", "2740143": "All segments are scored. It's good to keep in mind that vocalizations can overlap; we consider each species output as an independent binary classification problem, as each window may have no birds, one bird, or many birds." }, "source": "meta" }