{ "id": 238246, "title": "Last 3 weeks: What should you do?", "url": "/competitions/birdclef-2021/discussion/238246", "author_name": "Kamal Das", "post_date": "2021-05-11T16:52:17.120000", "votes": 17, "comment_count": 4, "views": 0, "content": "

Ok… been focusing more on HPA and Indoor and have not submitted anything in over 2 weeks..

\n

now back to see if we can create some magic and improve on the score here

\n

What should you do?

\n

1) Earlier comps
\nAs CPMP mentioned, look at earlier competitions and study them
\nlink: https://www.kaggle.com/c/birdclef-2021/discussion/234464

\n

Tensor girl as always has a list that you can refer to:
\nhttps://www.kaggle.com/c/birdclef-2021/discussion/230029
\n& by Sirish here: https://www.kaggle.com/c/birdclef-2021/discussion/229998

\n

Also some research papers https://www.kaggle.com/c/birdclef-2021/discussion/230025 (Thanks @usharengaraju)
\nand by Charlie Craine : https://www.kaggle.com/c/birdclef-2021/discussion/230043

\n

2) Ensemble and how?
\n@kneroma has been kind to suggest how to succeed \" bagging is the key\"
\nhttps://www.kaggle.com/c/birdclef-2021/discussion/233454

\n

Please also refer to his wonderful notebooks:
\nTraining: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab
\nInference: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference
\npls also go through the comments in the notebooks and ask whet you do not understand.
\nkkiller is very helpful!

\n

How to ensemble is a great discussion by Nitin Datta: https://www.kaggle.com/c/birdclef-2021/discussion/237502
\nAnd as the ever helpful @cdeotte notes, initially \"focus on one model and make it as accurate as possible\" then try to ensemble. Don't jump into ensembling straight away!
\nHis notes on post-process from an earlier audio comp are likely to be helpful here (https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389)

\n

3) Read more!
\nIn the Indoor comp, @mamasinkgs gave me some of the best advice:
\n(a) EDA by yourself.
\n(b) Understand good kernels (regardless of their score).
\n(Source: https://www.kaggle.com/c/indoor-location-navigation/discussion/228796)
\npart (b) is important-- don't just look at the top scoring Kernels. Lots of other notebooks have great approaches you can leverage.

\n

4) Trust your CV and build your own pipeline
\nWhile I do not expect surprises and shakeup like in hubmap, Theo Viel's suggestion will be ever helpful: https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/238150

\n

Trust your CV and good luck in the final few weeks.
\nAs CPMP has suggested, the GMs are coming… and there will be a lot of great ideas in the coming weeks
\nand looking to learn and understand more about audio analysis, once we discuss the best approaches!!

\n

ALL THE BEST!!

", "messages": [ { "id": 1302669, "postDate": "2021-05-11T16:52:17.120Z", "content": "

Ok… been focusing more on HPA and Indoor and have not submitted anything in over 2 weeks..

\n

now back to see if we can create some magic and improve on the score here

\n

What should you do?

\n

1) Earlier comps
\nAs CPMP mentioned, look at earlier competitions and study them
\nlink: https://www.kaggle.com/c/birdclef-2021/discussion/234464

\n

Tensor girl as always has a list that you can refer to:
\nhttps://www.kaggle.com/c/birdclef-2021/discussion/230029
\n& by Sirish here: https://www.kaggle.com/c/birdclef-2021/discussion/229998

\n

Also some research papers https://www.kaggle.com/c/birdclef-2021/discussion/230025 (Thanks @usharengaraju)
\nand by Charlie Craine : https://www.kaggle.com/c/birdclef-2021/discussion/230043

\n

2) Ensemble and how?
\n@kneroma has been kind to suggest how to succeed \" bagging is the key\"
\nhttps://www.kaggle.com/c/birdclef-2021/discussion/233454

\n

Please also refer to his wonderful notebooks:
\nTraining: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab
\nInference: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference
\npls also go through the comments in the notebooks and ask whet you do not understand.
\nkkiller is very helpful!

\n

How to ensemble is a great discussion by Nitin Datta: https://www.kaggle.com/c/birdclef-2021/discussion/237502
\nAnd as the ever helpful @cdeotte notes, initially \"focus on one model and make it as accurate as possible\" then try to ensemble. Don't jump into ensembling straight away!
\nHis notes on post-process from an earlier audio comp are likely to be helpful here (https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389)

\n

3) Read more!
\nIn the Indoor comp, @mamasinkgs gave me some of the best advice:
\n(a) EDA by yourself.
\n(b) Understand good kernels (regardless of their score).
\n(Source: https://www.kaggle.com/c/indoor-location-navigation/discussion/228796)
\npart (b) is important-- don't just look at the top scoring Kernels. Lots of other notebooks have great approaches you can leverage.

\n

4) Trust your CV and build your own pipeline
\nWhile I do not expect surprises and shakeup like in hubmap, Theo Viel's suggestion will be ever helpful: https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/238150

\n

Trust your CV and good luck in the final few weeks.
\nAs CPMP has suggested, the GMs are coming… and there will be a lot of great ideas in the coming weeks
\nand looking to learn and understand more about audio analysis, once we discuss the best approaches!!

\n

ALL THE BEST!!

", "rawMarkdown": "Ok... been focusing more on HPA and Indoor and have not submitted anything in over 2 weeks..\n\nnow back to see if we can create some magic and improve on the score here\n\nWhat should you do?\n\n1) Earlier comps\nAs CPMP mentioned, look at earlier competitions and study them\nlink: https://www.kaggle.com/c/birdclef-2021/discussion/234464\n\nTensor girl as always has a list that you can refer to:\nhttps://www.kaggle.com/c/birdclef-2021/discussion/230029 \n& by Sirish here: https://www.kaggle.com/c/birdclef-2021/discussion/229998\n\nAlso some research papers https://www.kaggle.com/c/birdclef-2021/discussion/230025 (Thanks @usharengaraju)\nand by Charlie Craine : https://www.kaggle.com/c/birdclef-2021/discussion/230043\n\n2) Ensemble and how?\n@kneroma has been kind to suggest how to succeed \" bagging is the key\"\nhttps://www.kaggle.com/c/birdclef-2021/discussion/233454\n\nPlease also refer to his wonderful notebooks:\nTraining: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\nInference: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference\npls also go through the comments in the notebooks and ask whet you do not understand. \nkkiller is very helpful!\n\nHow to ensemble is a great discussion by Nitin Datta: https://www.kaggle.com/c/birdclef-2021/discussion/237502\nAnd as the ever helpful @cdeotte notes, initially \"focus on one model and make it as accurate as possible\" then try to ensemble. Don't jump into ensembling straight away!\nHis notes on post-process from an earlier audio comp are likely to be helpful here (https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389)\n\n3) Read more!\nIn the Indoor comp, @mamasinkgs gave me some of the best advice:\n(a) EDA by yourself.\n(b) Understand good kernels (regardless of their score).\n(Source: https://www.kaggle.com/c/indoor-location-navigation/discussion/228796)\npart (b) is important-- don't just look at the top scoring Kernels. Lots of other notebooks have great approaches you can leverage. \n\n4) Trust your CV and build your own pipeline\nWhile I do not expect surprises and shakeup like in hubmap, Theo Viel's suggestion will be ever helpful: https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/238150\n\nTrust your CV and good luck in the final few weeks.\nAs CPMP has suggested, the GMs are coming... and there will be a lot of great ideas in the coming weeks\nand looking to learn and understand more about audio analysis, once we discuss the best approaches!!\n\nALL THE BEST!!\n\n\n", "votes": 17 }, { "id": 1302684, "postDate": "2021-05-11T16:57:55.880Z", "content": "
\n

Last 3 weeks: What should you do?

\n
\n

Let's save time for all and stop the competition right now.

", "rawMarkdown": "> Last 3 weeks: What should you do?\n\nLet's save time for all and stop the competition right now.", "votes": 12 }, { "id": 1315591, "postDate": "2021-05-20T00:57:17.093Z", "content": "

super helpfulll for people like me..

", "rawMarkdown": "super helpfulll for people like me..", "votes": 1 }, { "id": 1309085, "postDate": "2021-05-15T16:48:10.927Z", "content": "

@kmldas Thanks for post . Super Helpful for late starters like me

", "rawMarkdown": "@kmldas Thanks for post . Super Helpful for late starters like me", "votes": 2, "replies": [ { "id": 1309089, "postDate": "2021-05-15T16:52:12.920Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 1302684, "author_name": "CPMP", "author_url": "", "post_date": "2021-05-11T16:57:55.880000", "content": "
\n

Last 3 weeks: What should you do?

\n
\n

Let's save time for all and stop the competition right now.

", "votes": 12, "replies": [] }, { "id": 1315591, "author_name": "madara1234567", "author_url": "", "post_date": "2021-05-20T00:57:17.093000", "content": "

super helpfulll for people like me..

", "votes": 1, "replies": [] }, { "id": 1309085, "author_name": "Tensor Girl", "author_url": "", "post_date": "2021-05-15T16:48:10.927000", "content": "

@kmldas Thanks for post . Super Helpful for late starters like me

", "votes": 2, "replies": [ { "id": 1309089, "author_name": "", "author_url": "", "post_date": "2021-05-15T16:52:12.920000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1302669": "Ok... been focusing more on HPA and Indoor and have not submitted anything in over 2 weeks..\n\nnow back to see if we can create some magic and improve on the score here\n\nWhat should you do?\n\n1) Earlier comps\nAs CPMP mentioned, look at earlier competitions and study them\nlink: https://www.kaggle.com/c/birdclef-2021/discussion/234464\n\nTensor girl as always has a list that you can refer to:\nhttps://www.kaggle.com/c/birdclef-2021/discussion/230029 \n& by Sirish here: https://www.kaggle.com/c/birdclef-2021/discussion/229998\n\nAlso some research papers https://www.kaggle.com/c/birdclef-2021/discussion/230025 (Thanks @usharengaraju)\nand by Charlie Craine : https://www.kaggle.com/c/birdclef-2021/discussion/230043\n\n2) Ensemble and how?\n@kneroma has been kind to suggest how to succeed \" bagging is the key\"\nhttps://www.kaggle.com/c/birdclef-2021/discussion/233454\n\nPlease also refer to his wonderful notebooks:\nTraining: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\nInference: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference\npls also go through the comments in the notebooks and ask whet you do not understand. \nkkiller is very helpful!\n\nHow to ensemble is a great discussion by Nitin Datta: https://www.kaggle.com/c/birdclef-2021/discussion/237502\nAnd as the ever helpful @cdeotte notes, initially \"focus on one model and make it as accurate as possible\" then try to ensemble. Don't jump into ensembling straight away!\nHis notes on post-process from an earlier audio comp are likely to be helpful here (https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389)\n\n3) Read more!\nIn the Indoor comp, @mamasinkgs gave me some of the best advice:\n(a) EDA by yourself.\n(b) Understand good kernels (regardless of their score).\n(Source: https://www.kaggle.com/c/indoor-location-navigation/discussion/228796)\npart (b) is important-- don't just look at the top scoring Kernels. Lots of other notebooks have great approaches you can leverage. \n\n4) Trust your CV and build your own pipeline\nWhile I do not expect surprises and shakeup like in hubmap, Theo Viel's suggestion will be ever helpful: https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/238150\n\nTrust your CV and good luck in the final few weeks.\nAs CPMP has suggested, the GMs are coming... and there will be a lot of great ideas in the coming weeks\nand looking to learn and understand more about audio analysis, once we discuss the best approaches!!\n\nALL THE BEST!!\n\n\n", "1302684": "> Last 3 weeks: What should you do?\n\nLet's save time for all and stop the competition right now.", "1315591": "super helpfulll for people like me..", "1309085": "@kmldas Thanks for post . Super Helpful for late starters like me" } }