This looks like a tough competition. After a few weeks, the best leaderboard is only slightly better than the sample_submission.csv. When this competition ends, will the best leaderboard be similar to sample_submission.csv, or will the best leaderboard be better? What are your thoughts?
This looks like a tough competition. After a few weeks, the best leaderboard is only slightly better than the sample_submission.csv. When this competition ends, will the best leaderboard be similar to sample_submission.csv, or will the best leaderboard be better? What are your thoughts?
It will be better for sure (especially if you decide to join this competition 😄 )
", "rawMarkdown": "It will be better for sure (especially if you decide to join this competition 😄 ) ", "votes": 4 }, { "id": 971657, "postDate": "2020-08-15T19:23:15.323Z", "content": "My guess is 0.66 on public LB :)
", "rawMarkdown": "My guess is 0.66 on public LB :)", "votes": 1 }, { "id": 911755, "postDate": "2020-07-02T02:13:35.187Z", "content": "@cdeotte, in my opinion huge number of classes and mp3 data with all files with different bit rate and different duration.. this will surely be a hard nut to crack. However, I am new to kaggle. I was looking for a similar competition and found Freesound General-Purpose Audio Tagging Challenge. I saw a massive improvement over the sample submission.
\n\nYou can direct me where I can learn audio processing for ML. Thanks..
", "rawMarkdown": "@cdeotte, in my opinion huge number of classes and mp3 data with all files with different bit rate and different duration.. this will surely be a hard nut to crack. However, I am new to kaggle. I was looking for a similar competition and found [Freesound General-Purpose Audio Tagging Challenge](https://www.kaggle.com/c/freesound-audio-tagging). I saw a massive improvement over the sample submission. \n\nYou can direct me where I can learn audio processing for ML. Thanks.. ", "votes": 1, "replies": [ { "id": 914170, "postDate": "2020-07-03T16:55:39.707Z", "content": "@redwankarimsony Audio processing for music applications by Xavier sierra,is good place to learn about transforms and audio processing.
", "rawMarkdown": "@redwankarimsony Audio processing for music applications by Xavier sierra,is good place to learn about transforms and audio processing.", "votes": 1 }, { "id": 914503, "postDate": "2020-07-04T01:43:49.883Z", "content": "Thanks a lot @ashkhagan for the suggestion. My university is providing 9000 coursera courses free. I can readily join there :)
", "rawMarkdown": "Thanks a lot @ashkhagan for the suggestion. My university is providing 9000 coursera courses free. I can readily join there :) ", "votes": 1 } ] }, { "id": 914409, "postDate": "2020-07-03T20:58:08.387Z", "content": "I believe we will see a break through soon and some team will achieve LB 0.60+ !
", "rawMarkdown": "I believe we will see a break through soon and some team will achieve LB 0.60+ !", "votes": 2, "replies": [ { "id": 968465, "postDate": "2020-08-13T03:35:31.067Z", "content": "Someone surpassed LB 0.60!
", "rawMarkdown": "Someone surpassed LB 0.60!", "votes": 1 }, { "id": 968748, "postDate": "2020-08-13T08:11:00.143Z", "content": "it was very difficult
", "rawMarkdown": "it was very difficult", "votes": 3 }, { "id": 969774, "postDate": "2020-08-13T23:43:16.700Z", "content": "@vlomme Fantastic work!
", "rawMarkdown": "@vlomme Fantastic work!" } ] }, { "id": 913122, "postDate": "2020-07-03T02:05:59.287Z", "content": "my guess.. near 70s 😄
", "rawMarkdown": "my guess.. near 70s 😄 ", "votes": 2 }, { "id": 912088, "postDate": "2020-07-02T07:57:57.347Z", "content": "This competition will be primarily about pre-processing and the domain is, at least for me, a brand new one. Surely, pretty much everyone knows about fft and probably in most cases this is where the knowledge stops. Definitely not enough to progress here :) So the scores will go up, once enough people get on boarded and also the pipelines are built
", "rawMarkdown": "This competition will be primarily about pre-processing and the domain is, at least for me, a brand new one. Surely, pretty much everyone knows about fft and probably in most cases this is where the knowledge stops. Definitely not enough to progress here :) So the scores will go up, once enough people get on boarded and also the pipelines are built", "votes": 2, "replies": [ { "id": 914166, "postDate": "2020-07-03T16:54:18.343Z", "content": "got to bring in the big guns dft,stft then 😂
", "rawMarkdown": "got to bring in the big guns dft,stft then 😂 " }, { "id": 914501, "postDate": "2020-07-04T01:43:02.417Z", "content": "@ashkhagan, In my view, some of the factors like unequal audio duration, variable bit-rates, mute portion of the raw audio, unequal samples per class demands heavy pre processing even before applying DFT or STFT.
", "rawMarkdown": "@ashkhagan, In my view, some of the factors like unequal audio duration, variable bit-rates, mute portion of the raw audio, unequal samples per class demands heavy pre processing even before applying DFT or STFT. ", "votes": 2 } ] }, { "id": 914401, "postDate": "2020-07-03T20:45:44.427Z", "content": "Why is there a large group at 0.56?
", "rawMarkdown": "Why is there a large group at 0.56?", "replies": [ { "id": 914408, "postDate": "2020-07-03T20:56:38.280Z", "content": "Because everyone has hit a ceiling. The sample submission scores 0.54 and the best public notebook scores 0.56. So far we have not done much better than random guessing.
", "rawMarkdown": "Because everyone has hit a ceiling. The sample submission scores 0.54 and the best public notebook scores 0.56. So far we have not done much better than random guessing.", "votes": 3 }, { "id": 914418, "postDate": "2020-07-03T21:06:39.823Z", "content": "Is this a bad sign of potential shakeup?
", "rawMarkdown": "Is this a bad sign of potential shakeup?" } ] }, { "id": 914146, "postDate": "2020-07-03T16:38:59.290Z", "content": "I believe that the main issue with audio competitions not being so successful, like NLP and CV, is that there is no way of effectively applying transfer learning.
", "rawMarkdown": "I believe that the main issue with audio competitions not being so successful, like NLP and CV, is that there is no way of effectively applying transfer learning.", "replies": [ { "id": 914180, "postDate": "2020-07-03T17:00:41.087Z", "content": "Maybe if we can translate the birdsong into text, then we can use BERT 😄
", "rawMarkdown": "Maybe if we can translate the birdsong into text, then we can use BERT 😄 ", "votes": 5 }, { "id": 914185, "postDate": "2020-07-03T17:02:46.190Z", "content": "I think Gilles @group16 started translating some birdsongs into text here
", "rawMarkdown": "I think Gilles @group16 started translating some birdsongs into text [here][1]\n\n[1]: https://www.kaggle.com/c/birdsong-recognition/discussion/159167", "votes": 1 }, { "id": 914206, "postDate": "2020-07-03T17:15:31.570Z", "content": "Haha! Jokes aside, transformers is legitimate idea.
", "rawMarkdown": "Haha! Jokes aside, transformers is legitimate idea." }, { "id": 914620, "postDate": "2020-07-04T05:14:44.593Z", "content": "Don't underestimate my bird-speak Chris!
", "rawMarkdown": "Don't underestimate my bird-speak Chris!", "votes": 1 } ] }, { "id": 913088, "postDate": "2020-07-03T00:25:56.430Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 911794, "author_name": "yukiya", "author_url": "", "post_date": "2020-07-02T02:59:06.063000", "content": "It will be better for sure (especially if you decide to join this competition 😄 )
", "votes": 4, "replies": [] }, { "id": 971657, "author_name": "beluga", "author_url": "", "post_date": "2020-08-15T19:23:15.323000", "content": "My guess is 0.66 on public LB :)
", "votes": 1, "replies": [] }, { "id": 911755, "author_name": "Redwan Sony", "author_url": "", "post_date": "2020-07-02T02:13:35.187000", "content": "@cdeotte, in my opinion huge number of classes and mp3 data with all files with different bit rate and different duration.. this will surely be a hard nut to crack. However, I am new to kaggle. I was looking for a similar competition and found Freesound General-Purpose Audio Tagging Challenge. I saw a massive improvement over the sample submission.
\n\nYou can direct me where I can learn audio processing for ML. Thanks..
", "votes": 1, "replies": [ { "id": 914170, "author_name": "Ashwani Rathee", "author_url": "", "post_date": "2020-07-03T16:55:39.707000", "content": "@redwankarimsony Audio processing for music applications by Xavier sierra,is good place to learn about transforms and audio processing.
", "votes": 1, "replies": [] }, { "id": 914503, "author_name": "Redwan Sony", "author_url": "", "post_date": "2020-07-04T01:43:49.883000", "content": "Thanks a lot @ashkhagan for the suggestion. My university is providing 9000 coursera courses free. I can readily join there :)
", "votes": 1, "replies": [] } ] }, { "id": 914409, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-07-03T20:58:08.387000", "content": "I believe we will see a break through soon and some team will achieve LB 0.60+ !
", "votes": 2, "replies": [ { "id": 968465, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-08-13T03:35:31.067000", "content": "Someone surpassed LB 0.60!
", "votes": 1, "replies": [] }, { "id": 968748, "author_name": "Kramarenko Vladislav", "author_url": "", "post_date": "2020-08-13T08:11:00.143000", "content": "it was very difficult
", "votes": 3, "replies": [] }, { "id": 969774, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-08-13T23:43:16.700000", "content": "@vlomme Fantastic work!
", "votes": 0, "replies": [] } ] }, { "id": 913122, "author_name": "johnny", "author_url": "", "post_date": "2020-07-03T02:05:59.287000", "content": "my guess.. near 70s 😄
", "votes": 2, "replies": [] }, { "id": 912088, "author_name": "snovik", "author_url": "", "post_date": "2020-07-02T07:57:57.347000", "content": "This competition will be primarily about pre-processing and the domain is, at least for me, a brand new one. Surely, pretty much everyone knows about fft and probably in most cases this is where the knowledge stops. Definitely not enough to progress here :) So the scores will go up, once enough people get on boarded and also the pipelines are built
", "votes": 2, "replies": [ { "id": 914166, "author_name": "Ashwani Rathee", "author_url": "", "post_date": "2020-07-03T16:54:18.343000", "content": "got to bring in the big guns dft,stft then 😂
", "votes": 0, "replies": [] }, { "id": 914501, "author_name": "Redwan Sony", "author_url": "", "post_date": "2020-07-04T01:43:02.417000", "content": "@ashkhagan, In my view, some of the factors like unequal audio duration, variable bit-rates, mute portion of the raw audio, unequal samples per class demands heavy pre processing even before applying DFT or STFT.
", "votes": 2, "replies": [] } ] }, { "id": 914401, "author_name": "Bo Peng", "author_url": "", "post_date": "2020-07-03T20:45:44.427000", "content": "Why is there a large group at 0.56?
", "votes": 0, "replies": [ { "id": 914408, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-07-03T20:56:38.280000", "content": "Because everyone has hit a ceiling. The sample submission scores 0.54 and the best public notebook scores 0.56. So far we have not done much better than random guessing.
", "votes": 3, "replies": [] }, { "id": 914418, "author_name": "Bo Peng", "author_url": "", "post_date": "2020-07-03T21:06:39.823000", "content": "Is this a bad sign of potential shakeup?
", "votes": 0, "replies": [] } ] }, { "id": 914146, "author_name": "Nikola Bacic", "author_url": "", "post_date": "2020-07-03T16:38:59.290000", "content": "I believe that the main issue with audio competitions not being so successful, like NLP and CV, is that there is no way of effectively applying transfer learning.
", "votes": 0, "replies": [ { "id": 914180, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-07-03T17:00:41.087000", "content": "Maybe if we can translate the birdsong into text, then we can use BERT 😄
", "votes": 5, "replies": [] }, { "id": 914185, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-07-03T17:02:46.190000", "content": "I think Gilles @group16 started translating some birdsongs into text here
", "votes": 1, "replies": [] }, { "id": 914206, "author_name": "Nikola Bacic", "author_url": "", "post_date": "2020-07-03T17:15:31.570000", "content": "Haha! Jokes aside, transformers is legitimate idea.
", "votes": 0, "replies": [] }, { "id": 914620, "author_name": "Gilles Vandewiele", "author_url": "", "post_date": "2020-07-04T05:14:44.593000", "content": "Don't underestimate my bird-speak Chris!
", "votes": 1, "replies": [] } ] }, { "id": 913088, "author_name": "", "author_url": "", "post_date": "2020-07-03T00:25:56.430000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "911740": "This looks like a tough competition. After a few weeks, the best leaderboard is only slightly better than the `sample_submission.csv`. When this competition ends, will the best leaderboard be similar to `sample_submission.csv`, or will the best leaderboard be better? What are your thoughts?", "911794": "It will be better for sure (especially if you decide to join this competition 😄 ) ", "971657": "My guess is 0.66 on public LB :)", "911755": "@cdeotte, in my opinion huge number of classes and mp3 data with all files with different bit rate and different duration.. this will surely be a hard nut to crack. However, I am new to kaggle. I was looking for a similar competition and found [Freesound General-Purpose Audio Tagging Challenge](https://www.kaggle.com/c/freesound-audio-tagging). I saw a massive improvement over the sample submission. \n\nYou can direct me where I can learn audio processing for ML. Thanks.. ", "914409": "I believe we will see a break through soon and some team will achieve LB 0.60+ !", "913122": "my guess.. near 70s 😄 ", "912088": "This competition will be primarily about pre-processing and the domain is, at least for me, a brand new one. Surely, pretty much everyone knows about fft and probably in most cases this is where the knowledge stops. Definitely not enough to progress here :) So the scores will go up, once enough people get on boarded and also the pipelines are built", "914401": "Why is there a large group at 0.56?", "914146": "I believe that the main issue with audio competitions not being so successful, like NLP and CV, is that there is no way of effectively applying transfer learning.", "913088": "" } }