I've already ordered pizza and popcorn for reading the top solutions :)
", "messages": [ { "id": 1201867, "postDate": "2021-02-15T17:44:00.597Z", "content": "I've already ordered pizza and popcorn for reading the top solutions :)
", "rawMarkdown": "I've already ordered pizza and popcorn for reading the top solutions :)", "votes": 30 }, { "id": 1202012, "postDate": "2021-02-15T19:52:06.110Z", "content": "I didn't think I would find myself in a position to make one of those posts. But I spent a long time thinking about how to approach this competition and I'm glad to have good results, even if I was late and didn't get to make many submissions. If I'm in the final top 10 and no-one else's approach looks like mine, I'll do a write-up!
", "rawMarkdown": "I didn't think I would find myself in a position to make one of those posts. But I spent a long time thinking about how to approach this competition and I'm glad to have good results, even if I was late and didn't get to make many submissions. If I'm in the final top 10 and no-one else's approach looks like mine, I'll do a write-up!", "votes": 12, "replies": [ { "id": 1202120, "postDate": "2021-02-15T21:18:08.867Z", "content": "It doesn't matter what is the outcome you should still do write up =)
", "rawMarkdown": "It doesn't matter what is the outcome you should still do write up =) ", "votes": 11 }, { "id": 1204151, "postDate": "2021-02-15T22:09:09.830Z", "content": "Yeah, I definitely appreciate all of the write-ups regardless of final outcome. It is good to be able to do some sort of meta analysis and see the different techniques and how they influenced score
", "rawMarkdown": "Yeah, I definitely appreciate all of the write-ups regardless of final outcome. It is good to be able to do some sort of meta analysis and see the different techniques and how they influenced score", "votes": 6 }, { "id": 1204453, "postDate": "2021-02-16T06:42:35.207Z", "content": "being top 11th with less than 30 submission means that your have discovered a general good modeling framework.
\nJust make sure that your solution is robust and can withstand shakeup and you should be able to stay at the top.
\nNote that the tops are making more 100 to 300 submissions. They may have many models for ensembling to improve generalisation.
", "rawMarkdown": "being top 11th with less than 30 submission means that your have discovered a general good modeling framework.\n\nJust make sure that your solution is robust and can withstand shakeup and you should be able to stay at the top. \n\nNote that the tops are making more 100 to 300 submissions. They may have many models for ensembling to improve generalisation.", "votes": 1 }, { "id": 1204519, "postDate": "2021-02-16T07:14:18.423Z", "content": "ur approach is unique… as far as I can recall, I don't see a top 100 (let alone top 12) with less than 30 odd submissions…
", "rawMarkdown": "ur approach is unique... as far as I can recall, I don't see a top 100 (let alone top 12) with less than 30 odd submissions..." }, { "id": 1204842, "postDate": "2021-02-16T11:27:09.830Z", "content": "\n\nThey may have many models for ensembling to improve generalisation.
\n
actually I got 0.960+ with less than 30 subs with a single 5 fold model. The problem is that improving from 0.970 takes much more time than reaching it.
", "rawMarkdown": "> They may have many models for ensembling to improve generalisation.\n\nactually I got 0.960+ with less than 30 subs with a single 5 fold model. The problem is that improving from 0.970 takes much more time than reaching it. ", "votes": 11 }, { "id": 1204853, "postDate": "2021-02-16T11:29:30.460Z", "content": "great work
", "rawMarkdown": "great work\n", "votes": 1 }, { "id": 1205023, "postDate": "2021-02-16T13:40:06.393Z", "content": "i think the host should provide some full clip labels for kagglers to use for validation.
\nWithout clip-level labels, it is hard to do the validation.
Also, if the objective is to develop a model that can be generalized, downstream transfer learning to few-shot application may be a better approach. (something like GPT-3)
", "rawMarkdown": "i think the host should provide some full clip labels for kagglers to use for validation.\nWithout clip-level labels, it is hard to do the validation.\n\nAlso, if the objective is to develop a model that can be generalized, downstream transfer learning to few-shot application may be a better approach. (something like GPT-3)\n", "votes": 2 }, { "id": 1206596, "postDate": "2021-02-17T13:10:20.670Z", "content": "@bigironsphere You should write what you have done, regardless of what others did. Chances that someone else did exactly the same are negligible.
\nI will certainly share all the dead ends I explored ;) Maybe I should have stopped after my first sub as was suggested.
", "rawMarkdown": "@bigironsphere You should write what you have done, regardless of what others did. Chances that someone else did exactly the same are negligible.\n\nI will certainly share all the dead ends I explored ;) Maybe I should have stopped after my first sub as was suggested.", "votes": 4 }, { "id": 1206648, "postDate": "2021-02-17T13:45:11.487Z", "content": "@cpmp You're quite right, although I can't claim to have done anything particularly clever. My only real 'breakthrough' was how to incorporate the False Positive data - although it was likely common knowledge for more experienced competitors. I'll prepare a summary while my final model trains!
", "rawMarkdown": "@cpmp You're quite right, although I can't claim to have done anything particularly clever. My only real 'breakthrough' was how to incorporate the False Positive data - although it was likely common knowledge for more experienced competitors. I'll prepare a summary while my final model trains!", "votes": 2 }, { "id": 1206677, "postDate": "2021-02-17T13:55:49.750Z", "content": "Don't share anything before deadline though ;)
", "rawMarkdown": "Don't share anything before deadline though ;)", "votes": 1 }, { "id": 1206699, "postDate": "2021-02-17T14:02:47.007Z", "content": "\"My only real 'breakthrough' was how to incorporate the False Positive data\"
\nwow….cant wait to read about it…I did use FP but failed miserably :( the scores actually went a tad lower..
", "rawMarkdown": "*\"My only real 'breakthrough' was how to incorporate the False Positive data\"*\n\nwow....cant wait to read about it...I did use FP but failed miserably :( the scores actually went a tad lower.." } ] }, { "id": 1207895, "postDate": "2021-02-18T04:53:25.663Z", "content": "…I forgot to select the best submissions for the challenge. I fall from the 42th to 700th position lol
", "rawMarkdown": "...I forgot to select the best submissions for the challenge. I fall from the 42th to 700th position lol", "votes": 8, "replies": [ { "id": 1208487, "postDate": "2021-02-18T09:52:08.120Z", "content": "I got a good lesson from your mistake, friend 😍
", "rawMarkdown": "I got a good lesson from your mistake, friend 😍", "votes": 1 }, { "id": 1208585, "postDate": "2021-02-18T10:50:13.977Z", "content": "but Kaggle automatically selects top 2 submissions right? u might int to raise a bug report
", "rawMarkdown": "but Kaggle automatically selects top 2 submissions right? u might int to raise a bug report" }, { "id": 1208600, "postDate": "2021-02-18T11:03:05.117Z", "content": "In my case, Kaggle hasn't selected the top 2. Maybe I selected something at the beginning and then I forgot to update the best ones. Wouldn't be better if Kaggle asks, just before the end of the competition, which submission to use? …like confirmation as a double check? I know it's my fault, maybe…
", "rawMarkdown": "In my case, Kaggle hasn't selected the top 2. Maybe I selected something at the beginning and then I forgot to update the best ones. Wouldn't be better if Kaggle asks, just before the end of the competition, which submission to use? ...like confirmation as a double check? I know it's my fault, maybe..." }, { "id": 1209355, "postDate": "2021-02-18T21:06:56.153Z", "content": "Oh!
\nI'm the one who wrote to you right now, asking what has happened.. Now I know. Sorry to hear that.
Admit it, you wanted to show ther will be a huge shakeup.
\nJust kidding of course, this is bad. You probably forgot an early selection.
", "rawMarkdown": "Admit it, you wanted to show ther will be a huge shakeup.\n\nJust kidding of course, this is bad. You probably forgot an early selection." }, { "id": 1209854, "postDate": "2021-02-19T04:29:55.667Z", "content": "I feel really sorry for you. I wonder if you still want to share your CAM attention approach? I am sure a lot of people like me can learn from it.
", "rawMarkdown": "I feel really sorry for you. I wonder if you still want to share your CAM attention approach? I am sure a lot of people like me can learn from it.", "isDeleted": true } ] }, { "id": 1206553, "postDate": "2021-02-17T12:31:31.983Z", "content": "And I want to see the simplest one with 95% score. BTW, is there a place with the simplest solutions of competitions (who reached 95% of the best score)?
", "rawMarkdown": "And I want to see the simplest one with 95% score. BTW, is there a place with the simplest solutions of competitions (who reached 95% of the best score)?", "votes": 3 }, { "id": 1204261, "postDate": "2021-02-16T01:34:52.697Z", "content": "Yeah. Can't wait to learn about people's intelligent cropping and magical augmentations!
", "rawMarkdown": "Yeah. Can't wait to learn about people's intelligent cropping and magical augmentations!", "votes": 3, "replies": [ { "id": 1205000, "postDate": "2021-02-16T13:27:34.667Z", "content": "Good Luck you all for this Milkshake or LBshake or whatever !🙌
", "rawMarkdown": "Good Luck you all for this Milkshake or LBshake or whatever !🙌", "votes": 1 }, { "id": 1207229, "postDate": "2021-02-17T18:48:47.453Z", "content": "Didn't participate in the contest but after going through this thread, I have my pizza ready too… Cheers.
", "rawMarkdown": "Didn't participate in the contest but after going through this thread, I have my pizza ready too... Cheers.", "votes": 1, "replies": [ { "id": 1207234, "postDate": "2021-02-17T18:51:05.997Z", "content": "Pizza will be long cold when the competition ends 😨
", "rawMarkdown": "Pizza will be long cold when the competition ends 😨", "votes": 1 }, { "id": 1207238, "postDate": "2021-02-17T18:54:08.470Z", "content": "I am sure a sizzling notebook such as yours will warm it up 😉
", "rawMarkdown": "I am sure a sizzling notebook such as yours will warm it up 😉", "votes": 1 } ] }, { "id": 1205525, "postDate": "2021-02-16T18:32:11.743Z", "content": "Very curious to see how people approached this. I wasn't able to put a lot of time into this competition after the first few weeks but most of my time was spent looking into different approaches to working with label noise. If I had more time I would have spent it on cleaning labels and trying things like noise correction while training.
", "rawMarkdown": "Very curious to see how people approached this. I wasn't able to put a lot of time into this competition after the first few weeks but most of my time was spent looking into different approaches to working with label noise. If I had more time I would have spent it on cleaning labels and trying things like noise correction while training. ", "votes": 1 }, { "id": 1204993, "postDate": "2021-02-16T13:24:37.500Z", "content": "good job on top 50, That is really good in it's own right
", "rawMarkdown": "good job on top 50, That is really good in it's own right", "votes": 1 }, { "id": 1207517, "postDate": "2021-02-17T22:13:23.623Z", "content": "While we are waiting for the shakeup:
\nhttps://www.youtube.com/watch?v=jofNR_WkoCE
did you get the cheesy crust with garlic bread ?
", "rawMarkdown": "did you get the cheesy crust with garlic bread ?", "replies": [ { "id": 1211053, "postDate": "2021-02-19T23:26:25.027Z", "content": "Of course no! In Italy we have only great pizzas :)
", "rawMarkdown": "Of course no! In Italy we have only great pizzas :)", "votes": 2 } ] }, { "id": 1207431, "postDate": "2021-02-17T20:52:55.513Z", "content": "I'm curios about the FP data usage. Unfortunately, I couldn't manage it to work
", "rawMarkdown": "I'm curios about the FP data usage. Unfortunately, I couldn't manage it to work" }, { "id": 1205202, "postDate": "2021-02-16T15:38:22.183Z", "content": "yeah, I am curious about ensemble models and SED validation tricks trained on both tp and fp.
", "rawMarkdown": "yeah, I am curious about ensemble models and SED validation tricks trained on both tp and fp. " }, { "id": 1204501, "postDate": "2021-02-16T07:04:19.067Z", "content": "yes, can't wait to learn about people's intelligent cropping.
", "rawMarkdown": "yes, can't wait to learn about people's intelligent cropping.\n", "replies": [ { "id": 1204997, "postDate": "2021-02-16T13:27:04.673Z", "content": "I think it's just cropping the TPs around the positive signal
", "rawMarkdown": "I think it's just cropping the TPs around the positive signal" } ] }, { "id": 1204446, "postDate": "2021-02-16T06:29:58.003Z", "content": "yeah me too
", "rawMarkdown": "yeah me too" } ], "comments": [ { "id": 1202012, "author_name": "RNA", "author_url": "", "post_date": "2021-02-15T19:52:06.110000", "content": "I didn't think I would find myself in a position to make one of those posts. But I spent a long time thinking about how to approach this competition and I'm glad to have good results, even if I was late and didn't get to make many submissions. If I'm in the final top 10 and no-one else's approach looks like mine, I'll do a write-up!
", "votes": 12, "replies": [ { "id": 1202120, "author_name": "DrHB", "author_url": "", "post_date": "2021-02-15T21:18:08.867000", "content": "It doesn't matter what is the outcome you should still do write up =)
", "votes": 11, "replies": [] }, { "id": 1204151, "author_name": "ryches", "author_url": "", "post_date": "2021-02-15T22:09:09.830000", "content": "Yeah, I definitely appreciate all of the write-ups regardless of final outcome. It is good to be able to do some sort of meta analysis and see the different techniques and how they influenced score
", "votes": 6, "replies": [] }, { "id": 1204453, "author_name": "hengck23", "author_url": "", "post_date": "2021-02-16T06:42:35.207000", "content": "being top 11th with less than 30 submission means that your have discovered a general good modeling framework.
\nJust make sure that your solution is robust and can withstand shakeup and you should be able to stay at the top.
\nNote that the tops are making more 100 to 300 submissions. They may have many models for ensembling to improve generalisation.
", "votes": 1, "replies": [] }, { "id": 1204519, "author_name": "Allohvk", "author_url": "", "post_date": "2021-02-16T07:14:18.423000", "content": "ur approach is unique… as far as I can recall, I don't see a top 100 (let alone top 12) with less than 30 odd submissions…
", "votes": 0, "replies": [] }, { "id": 1204842, "author_name": "Selim Seferbekov", "author_url": "", "post_date": "2021-02-16T11:27:09.830000", "content": "\n\nThey may have many models for ensembling to improve generalisation.
\n
actually I got 0.960+ with less than 30 subs with a single 5 fold model. The problem is that improving from 0.970 takes much more time than reaching it.
", "votes": 11, "replies": [] }, { "id": 1204853, "author_name": "RIYADH HAITHAM", "author_url": "", "post_date": "2021-02-16T11:29:30.460000", "content": "great work
", "votes": 1, "replies": [] }, { "id": 1205023, "author_name": "hengck23", "author_url": "", "post_date": "2021-02-16T13:40:06.393000", "content": "i think the host should provide some full clip labels for kagglers to use for validation.
\nWithout clip-level labels, it is hard to do the validation.
Also, if the objective is to develop a model that can be generalized, downstream transfer learning to few-shot application may be a better approach. (something like GPT-3)
", "votes": 2, "replies": [] }, { "id": 1206596, "author_name": "CPMP", "author_url": "", "post_date": "2021-02-17T13:10:20.670000", "content": "@bigironsphere You should write what you have done, regardless of what others did. Chances that someone else did exactly the same are negligible.
\nI will certainly share all the dead ends I explored ;) Maybe I should have stopped after my first sub as was suggested.
", "votes": 4, "replies": [] }, { "id": 1206648, "author_name": "RNA", "author_url": "", "post_date": "2021-02-17T13:45:11.487000", "content": "@cpmp You're quite right, although I can't claim to have done anything particularly clever. My only real 'breakthrough' was how to incorporate the False Positive data - although it was likely common knowledge for more experienced competitors. I'll prepare a summary while my final model trains!
", "votes": 2, "replies": [] }, { "id": 1206677, "author_name": "CPMP", "author_url": "", "post_date": "2021-02-17T13:55:49.750000", "content": "Don't share anything before deadline though ;)
", "votes": 1, "replies": [] }, { "id": 1206699, "author_name": "Allohvk", "author_url": "", "post_date": "2021-02-17T14:02:47.007000", "content": "\"My only real 'breakthrough' was how to incorporate the False Positive data\"
\nwow….cant wait to read about it…I did use FP but failed miserably :( the scores actually went a tad lower..
", "votes": 0, "replies": [] } ] }, { "id": 1207895, "author_name": "Guglielmo Camporese", "author_url": "", "post_date": "2021-02-18T04:53:25.663000", "content": "…I forgot to select the best submissions for the challenge. I fall from the 42th to 700th position lol
", "votes": 8, "replies": [ { "id": 1208487, "author_name": "Rajkumar Lakshmanamoorthy", "author_url": "", "post_date": "2021-02-18T09:52:08.120000", "content": "I got a good lesson from your mistake, friend 😍
", "votes": 1, "replies": [] }, { "id": 1208585, "author_name": "Allohvk", "author_url": "", "post_date": "2021-02-18T10:50:13.977000", "content": "but Kaggle automatically selects top 2 submissions right? u might int to raise a bug report
", "votes": 0, "replies": [] }, { "id": 1208600, "author_name": "Guglielmo Camporese", "author_url": "", "post_date": "2021-02-18T11:03:05.117000", "content": "In my case, Kaggle hasn't selected the top 2. Maybe I selected something at the beginning and then I forgot to update the best ones. Wouldn't be better if Kaggle asks, just before the end of the competition, which submission to use? …like confirmation as a double check? I know it's my fault, maybe…
", "votes": 0, "replies": [] }, { "id": 1209355, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-02-18T21:06:56.153000", "content": "Oh!
\nI'm the one who wrote to you right now, asking what has happened.. Now I know. Sorry to hear that.
Admit it, you wanted to show ther will be a huge shakeup.
\nJust kidding of course, this is bad. You probably forgot an early selection.
", "votes": 0, "replies": [] }, { "id": 1209854, "author_name": "", "author_url": "", "post_date": "2021-02-19T04:29:55.667000", "content": "I feel really sorry for you. I wonder if you still want to share your CAM attention approach? I am sure a lot of people like me can learn from it.
", "votes": 0, "replies": [] } ] }, { "id": 1206553, "author_name": "Sergii Kryvenko", "author_url": "", "post_date": "2021-02-17T12:31:31.983000", "content": "And I want to see the simplest one with 95% score. BTW, is there a place with the simplest solutions of competitions (who reached 95% of the best score)?
", "votes": 3, "replies": [] }, { "id": 1204261, "author_name": "dan", "author_url": "", "post_date": "2021-02-16T01:34:52.697000", "content": "Yeah. Can't wait to learn about people's intelligent cropping and magical augmentations!
", "votes": 3, "replies": [ { "id": 1205000, "author_name": "Guglielmo Camporese", "author_url": "", "post_date": "2021-02-16T13:27:34.667000", "content": "Good Luck you all for this Milkshake or LBshake or whatever !🙌
", "votes": 1, "replies": [] }, { "id": 1207229, "author_name": "Aadarsh Singh", "author_url": "", "post_date": "2021-02-17T18:48:47.453000", "content": "Didn't participate in the contest but after going through this thread, I have my pizza ready too… Cheers.
", "votes": 1, "replies": [ { "id": 1207234, "author_name": "Pascal Pfeiffer", "author_url": "", "post_date": "2021-02-17T18:51:05.997000", "content": "Pizza will be long cold when the competition ends 😨
", "votes": 1, "replies": [] }, { "id": 1207238, "author_name": "Aadarsh Singh", "author_url": "", "post_date": "2021-02-17T18:54:08.470000", "content": "I am sure a sizzling notebook such as yours will warm it up 😉
", "votes": 1, "replies": [] } ] }, { "id": 1205525, "author_name": "Jack Vial", "author_url": "", "post_date": "2021-02-16T18:32:11.743000", "content": "Very curious to see how people approached this. I wasn't able to put a lot of time into this competition after the first few weeks but most of my time was spent looking into different approaches to working with label noise. If I had more time I would have spent it on cleaning labels and trying things like noise correction while training.
", "votes": 1, "replies": [] }, { "id": 1204993, "author_name": "Dom De Jonge", "author_url": "", "post_date": "2021-02-16T13:24:37.500000", "content": "good job on top 50, That is really good in it's own right
", "votes": 1, "replies": [] }, { "id": 1207517, "author_name": "beluga", "author_url": "", "post_date": "2021-02-17T22:13:23.623000", "content": "While we are waiting for the shakeup:
\nhttps://www.youtube.com/watch?v=jofNR_WkoCE
did you get the cheesy crust with garlic bread ?
", "votes": 0, "replies": [ { "id": 1211053, "author_name": "Guglielmo Camporese", "author_url": "", "post_date": "2021-02-19T23:26:25.027000", "content": "Of course no! In Italy we have only great pizzas :)
", "votes": 2, "replies": [] } ] }, { "id": 1207431, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2021-02-17T20:52:55.513000", "content": "I'm curios about the FP data usage. Unfortunately, I couldn't manage it to work
", "votes": 0, "replies": [] }, { "id": 1205202, "author_name": "Neo", "author_url": "", "post_date": "2021-02-16T15:38:22.183000", "content": "yeah, I am curious about ensemble models and SED validation tricks trained on both tp and fp.
", "votes": 0, "replies": [] }, { "id": 1204501, "author_name": "Muhammad Ali", "author_url": "", "post_date": "2021-02-16T07:04:19.067000", "content": "yes, can't wait to learn about people's intelligent cropping.
", "votes": 0, "replies": [ { "id": 1204997, "author_name": "Guglielmo Camporese", "author_url": "", "post_date": "2021-02-16T13:27:04.673000", "content": "I think it's just cropping the TPs around the positive signal
", "votes": 0, "replies": [] } ] }, { "id": 1204446, "author_name": "RIYADH HAITHAM", "author_url": "", "post_date": "2021-02-16T06:29:58.003000", "content": "yeah me too
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1201867": "I've already ordered pizza and popcorn for reading the top solutions :)", "1202012": "I didn't think I would find myself in a position to make one of those posts. But I spent a long time thinking about how to approach this competition and I'm glad to have good results, even if I was late and didn't get to make many submissions. If I'm in the final top 10 and no-one else's approach looks like mine, I'll do a write-up!", "1207895": "...I forgot to select the best submissions for the challenge. I fall from the 42th to 700th position lol", "1206553": "And I want to see the simplest one with 95% score. BTW, is there a place with the simplest solutions of competitions (who reached 95% of the best score)?", "1204261": "Yeah. Can't wait to learn about people's intelligent cropping and magical augmentations!", "1207585": "Good Luck you all for this Milkshake or LBshake or whatever !🙌", "1207229": "Didn't participate in the contest but after going through this thread, I have my pizza ready too... Cheers.", "1205525": "Very curious to see how people approached this. I wasn't able to put a lot of time into this competition after the first few weeks but most of my time was spent looking into different approaches to working with label noise. If I had more time I would have spent it on cleaning labels and trying things like noise correction while training. ", "1204993": "good job on top 50, That is really good in it's own right", "1207517": "While we are waiting for the shakeup:\nhttps://www.youtube.com/watch?v=jofNR_WkoCE", "1211048": "did you get the cheesy crust with garlic bread ?", "1207431": "I'm curios about the FP data usage. Unfortunately, I couldn't manage it to work", "1205202": "yeah, I am curious about ensemble models and SED validation tricks trained on both tp and fp. ", "1204501": "yes, can't wait to learn about people's intelligent cropping.\n", "1204446": "yeah me too" } }