{ "id": 216870, "title": "Which metric are you using?", "url": "/competitions/rfcx-species-audio-detection/discussion/216870", "author_name": "", "post_date": "2021-02-04T10:49:26.961477100Z", "votes": 5, "comment_count": 1, "views": 0, "content": "

Hi,
\nI am using LWLRAP(the competition metric) and though my local CV is 0.89+ my highest LB score is 0.857 (I got 0.880 by ensembling couple of submissions). I found most of the notebooks using this competition metric but some(like @barnwellguy )are using AUC, precision/recall

\n

Which metric are you using?

\n

I am just a beginner…all answers are appreciated.
\nThank you 🙏

", "messages": [ { "id": "1185748", "postDate": "02/04/2021 10:49:26", "content": "

Hi,
\nI am using LWLRAP(the competition metric) and though my local CV is 0.89+ my highest LB score is 0.857 (I got 0.880 by ensembling couple of submissions). I found most of the notebooks using this competition metric but some(like @barnwellguy )are using AUC, precision/recall

\n

Which metric are you using?

\n

I am just a beginner…all answers are appreciated.
\nThank you 🙏

", "rawMarkdown": "Hi,\nI am using LWLRAP(the competition metric) and though my local CV is 0.89+ my highest LB score is 0.857 (I got 0.880 by ensembling couple of submissions). I found most of the notebooks using this competition metric but some(like @barnwellguy )are using AUC, precision/recall \n\n\nWhich metric are you using?\n\nI am just a beginner...all answers are appreciated.\nThank you 🙏", "votes": null }, { "id": "1185821", "postDate": "02/04/2021 11:34:35", "content": "

The gap between CV and LB is not surprising, because you cannot really get a correct LWLRAP on the CV due to only parts of each audio clip being annotated (this thread explains the labelling process and how it differed between the training data we have and the public/private LB test data). There's some answers to your questions from people near the top in this thread.

", "rawMarkdown": "The gap between CV and LB is not surprising, because you cannot really get a correct LWLRAP on the CV due to only parts of each audio clip being annotated ([this thread](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/206980) explains the labelling process and how it differed between the training data we have and the public/private LB test data). There's some answers to your questions from people near the top in [this thread](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/216596).", "votes": null } ], "comments": [ { "id": 1185821, "author_name": "bjoernholzhauer", "author_url": "", "post_date": "02/04/2021 11:34:35", "content": "

The gap between CV and LB is not surprising, because you cannot really get a correct LWLRAP on the CV due to only parts of each audio clip being annotated (this thread explains the labelling process and how it differed between the training data we have and the public/private LB test data). There's some answers to your questions from people near the top in this thread.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1185748": "Hi,\nI am using LWLRAP(the competition metric) and though my local CV is 0.89+ my highest LB score is 0.857 (I got 0.880 by ensembling couple of submissions). I found most of the notebooks using this competition metric but some(like @barnwellguy )are using AUC, precision/recall \n\n\nWhich metric are you using?\n\nI am just a beginner...all answers are appreciated.\nThank you 🙏", "1185821": "The gap between CV and LB is not surprising, because you cannot really get a correct LWLRAP on the CV due to only parts of each audio clip being annotated ([this thread](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/206980) explains the labelling process and how it differed between the training data we have and the public/private LB test data). There's some answers to your questions from people near the top in [this thread](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/216596)." }, "source": "meta" }