Will there be multiple birds speaking at the same time, and we're expected to output 0.5 for both? If not, wouldn't there at least be a case where one 5-second interval would have 2 birds sequentially calling? Am I misunderstanding, and this isn't actually a simple classification problem after all?
", "messages": [ { "id": "2744619", "postDate": "04/10/2024 03:35:47", "content": "Will there be multiple birds speaking at the same time, and we're expected to output 0.5 for both? If not, wouldn't there at least be a case where one 5-second interval would have 2 birds sequentially calling? Am I misunderstanding, and this isn't actually a simple classification problem after all?
", "rawMarkdown": "Will there be multiple birds speaking at the same time, and we're expected to output 0.5 for both? If not, wouldn't there at least be a case where one 5-second interval would have 2 birds sequentially calling? Am I misunderstanding, and this isn't actually a simple classification problem after all?", "votes": null }, { "id": "2745183", "postDate": "04/10/2024 12:53:59", "content": "\n\nIn multi-class classification, each input will have only one output class, but in multi-label classification, each input can have multi-output classes
\n
This isn't only a multi-class problem, but a multi-label problem. Your model should output 1 and 1 for both species present in the segment.
\nTo give you an example of how to change the modelling side of your pipeline, a multi class problem would typically use a softmax along the classes axis, whereas the multi label problem would use a sigmoid to let the model be able to predict anything between 0 and 1 for each class. Implying a different loss function, which I let you find.
\nAlso look at the secondary_label column present in the metadata.
I didn't know that, thanks.
\nAnother thing: are you expected to predict 1 for a bird call that appears, regardless of how long it is? For example if bird 1 calls for two seconds and bird 2 calls for 0.5 seconds, do you predict 1 for both? Thanks.
", "rawMarkdown": "I didn't know that, thanks. \n\nAnother thing: are you expected to predict 1 for a bird call that appears, regardless of how long it is? For example if bird 1 calls for two seconds and bird 2 calls for 0.5 seconds, do you predict 1 for both? Thanks.", "votes": null }, { "id": "2762724", "postDate": "04/20/2024 03:57:51", "content": "Sorry I missed your response, yes, if a bird appears in a 5sec segment for even 0.1sec, it should be labelled as 1
", "rawMarkdown": "Sorry I missed your response, yes, if a bird appears in a 5sec segment for even 0.1sec, it should be labelled as 1", "votes": null }, { "id": "2762784", "postDate": "04/20/2024 05:03:29", "content": "Ah ok, thanks.
", "rawMarkdown": "Ah ok, thanks.", "votes": null } ], "comments": [ { "id": 2745183, "author_name": "janmpia", "author_url": "", "post_date": "04/10/2024 12:53:59", "content": "\n\nIn multi-class classification, each input will have only one output class, but in multi-label classification, each input can have multi-output classes
\n
This isn't only a multi-class problem, but a multi-label problem. Your model should output 1 and 1 for both species present in the segment.
\nTo give you an example of how to change the modelling side of your pipeline, a multi class problem would typically use a softmax along the classes axis, whereas the multi label problem would use a sigmoid to let the model be able to predict anything between 0 and 1 for each class. Implying a different loss function, which I let you find.
\nAlso look at the secondary_label column present in the metadata.
I didn't know that, thanks.
\nAnother thing: are you expected to predict 1 for a bird call that appears, regardless of how long it is? For example if bird 1 calls for two seconds and bird 2 calls for 0.5 seconds, do you predict 1 for both? Thanks.
", "votes": null, "replies": [ { "id": 2762724, "author_name": "janmpia", "author_url": "", "post_date": "04/20/2024 03:57:51", "content": "Sorry I missed your response, yes, if a bird appears in a 5sec segment for even 0.1sec, it should be labelled as 1
", "votes": null, "replies": [ { "id": 2762784, "author_name": "rfeng12", "author_url": "", "post_date": "04/20/2024 05:03:29", "content": "Ah ok, thanks.
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "2744619": "Will there be multiple birds speaking at the same time, and we're expected to output 0.5 for both? If not, wouldn't there at least be a case where one 5-second interval would have 2 birds sequentially calling? Am I misunderstanding, and this isn't actually a simple classification problem after all?", "2745183": "> In multi-class classification, each input will have only one output class, but in multi-label classification, each input can have multi-output classes\n\nThis isn't only a multi-class problem, but a multi-label problem. Your model should output 1 and 1 for both species present in the segment. \nTo give you an example of how to change the modelling side of your pipeline, a multi class problem would typically use a softmax along the classes axis, whereas the multi label problem would use a sigmoid to let the model be able to predict anything between 0 and 1 for each class. Implying a different loss function, which I let you find. \nAlso look at the `secondary_label` column present in the metadata.", "2745648": "I didn't know that, thanks. \n\nAnother thing: are you expected to predict 1 for a bird call that appears, regardless of how long it is? For example if bird 1 calls for two seconds and bird 2 calls for 0.5 seconds, do you predict 1 for both? Thanks.", "2762724": "Sorry I missed your response, yes, if a bird appears in a 5sec segment for even 0.1sec, it should be labelled as 1", "2762784": "Ah ok, thanks." }, "source": "meta" }