Does this column reflect the audio quality? Will using this column as an objective in multi-task learning make the network more robust?🧐
", "messages": [ { "id": 2804846, "postDate": "2024-05-10T08:30:44.143Z", "content": "Does this column reflect the audio quality? Will using this column as an objective in multi-task learning make the network more robust?🧐
", "rawMarkdown": "Does this column reflect the audio quality? Will using this column as an objective in multi-task learning make the network more robust?🧐", "votes": 1 }, { "id": 2806069, "postDate": "2024-05-10T21:06:38.123Z", "content": "I tried to weight based off this rating but ultimately found no success with it. The CV score increases but the LB score falls off massively. I think it is just because of the quality comparison between the 2. As the competition leaders pointed out at one point. The CV is like you are listening for a bird in a room with only you and the bird. The test set is like standing in a garden and trying to listen for the bird.
", "rawMarkdown": "I tried to weight based off this rating but ultimately found no success with it. The CV score increases but the LB score falls off massively. I think it is just because of the quality comparison between the 2. As the competition leaders pointed out at one point. The CV is like you are listening for a bird in a room with only you and the bird. The test set is like standing in a garden and trying to listen for the bird. ", "votes": 2, "replies": [ { "id": 2806291, "postDate": "2024-05-11T02:41:59.767Z", "content": "Thanks Cody, but what's the meaning of 'weight based off the rating' ? does it mean the CV split strategy or mixup like weight? Sorry i am still a bit confused.
", "rawMarkdown": "Thanks Cody, but what's the meaning of 'weight based off the rating' ? does it mean the CV split strategy or mixup like weight? Sorry i am still a bit confused.", "votes": 1, "replies": [ { "id": 2806352, "postDate": "2024-05-11T03:56:55.240Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2806704, "postDate": "2024-05-11T08:22:44.590Z", "content": "Thanks Cody, I understand what you mean. Is the rating value marked by many different people? Or is it obtained by some objective rules? So if it is marked by different people, will it be less objective?
", "rawMarkdown": "Thanks Cody, I understand what you mean. Is the rating value marked by many different people? Or is it obtained by some objective rules? So if it is marked by different people, will it be less objective?" }, { "id": 2807024, "postDate": "2024-05-11T13:04:14.567Z", "content": "Since the rating is given by the Xeno-Canto community, it appears to be given by more than one name.
\nhttps://www.kaggle.com/competitions/birdclef-2024/discussion/491145
Also, the rating itself seems to indicate whether or not there is noise other than the primary label bird, not the sound quality.
\nhttps://www.kaggle.com/competitions/birdclef-2024/discussion/493605
In BirdCLEF2021, there is a top solution that uses mixup with rating as weight, so it may be a metadata that can increase the score if used well.
\nhttps://www.kaggle.com/competitions/birdclef-2021/discussion/243463
Thanks tc for your detailed reply!
", "rawMarkdown": "Thanks tc for your detailed reply!" } ] } ] } ], "comments": [ { "id": 2806069, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-05-10T21:06:38.123000", "content": "I tried to weight based off this rating but ultimately found no success with it. The CV score increases but the LB score falls off massively. I think it is just because of the quality comparison between the 2. As the competition leaders pointed out at one point. The CV is like you are listening for a bird in a room with only you and the bird. The test set is like standing in a garden and trying to listen for the bird.
", "votes": 2, "replies": [ { "id": 2806291, "author_name": "tanxxx", "author_url": "", "post_date": "2024-05-11T02:41:59.767000", "content": "Thanks Cody, but what's the meaning of 'weight based off the rating' ? does it mean the CV split strategy or mixup like weight? Sorry i am still a bit confused.
", "votes": 1, "replies": [ { "id": 2806352, "author_name": "", "author_url": "", "post_date": "2024-05-11T03:56:55.240000", "content": "", "votes": 0, "replies": [] }, { "id": 2806704, "author_name": "tanxxx", "author_url": "", "post_date": "2024-05-11T08:22:44.590000", "content": "Thanks Cody, I understand what you mean. Is the rating value marked by many different people? Or is it obtained by some objective rules? So if it is marked by different people, will it be less objective?
", "votes": 0, "replies": [] }, { "id": 2807024, "author_name": "tc0000", "author_url": "", "post_date": "2024-05-11T13:04:14.567000", "content": "Since the rating is given by the Xeno-Canto community, it appears to be given by more than one name.
\nhttps://www.kaggle.com/competitions/birdclef-2024/discussion/491145
Also, the rating itself seems to indicate whether or not there is noise other than the primary label bird, not the sound quality.
\nhttps://www.kaggle.com/competitions/birdclef-2024/discussion/493605
In BirdCLEF2021, there is a top solution that uses mixup with rating as weight, so it may be a metadata that can increase the score if used well.
\nhttps://www.kaggle.com/competitions/birdclef-2021/discussion/243463
Thanks tc for your detailed reply!
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2804846": "Does this column reflect the audio quality? Will using this column as an objective in multi-task learning make the network more robust?🧐", "2806069": "I tried to weight based off this rating but ultimately found no success with it. The CV score increases but the LB score falls off massively. I think it is just because of the quality comparison between the 2. As the competition leaders pointed out at one point. The CV is like you are listening for a bird in a room with only you and the bird. The test set is like standing in a garden and trying to listen for the bird. " } }