The best thing I learned from this competition is writing my own code. When I joined this competition I decided to write my own code and understand every line of it which lead to more research about terms, functions, methods, etc… Pytorch docs and competition's discussion were my friends during this competition, and I am very satisfied with my results (given that I couldn't grab a medal according to the public LB).
\nGood luck to all of you!
The best thing I learned from this competition is writing my own code. When I joined this competition I decided to write my own code and understand every line of it which lead to more research about terms, functions, methods, etc… Pytorch docs and competition's discussion were my friends during this competition, and I am very satisfied with my results (given that I couldn't grab a medal according to the public LB).
\nGood luck to all of you!
You're within 0.004 of a silver medal on the public LB. If you end up in the medals, that would not be a shocking shake-up, at all - especially, if you took a very different approach of your own with a strong validation strategy - rather than relying on the public LB score or just using the same public notebook with a 0.903 score that many people appear to have used. On the other hand, despite having some models of my own that I think I optimized properly via CV, I've peaked out at what gets rounded to 0.903, too, so perhaps that turned out to be a natural limit for a lot of us?! Despite seeing a medal on the public LB right now and (of course) hoping to not drop in the ranking, I'm aware that a drop by 1500 places on the private LB is something that should not be a complete shock given all the crowding just above 0.900.
", "rawMarkdown": "You're within 0.004 of a silver medal on the public LB. If you end up in the medals, that would not be a shocking shake-up, at all - especially, if you took a very different approach of your own with a strong validation strategy - rather than relying on the public LB score or just using the same public notebook with a 0.903 score that many people appear to have used. On the other hand, despite having some models of my own that I think I optimized properly via CV, I've peaked out at what gets rounded to 0.903, too, so perhaps that turned out to be a natural limit for a lot of us?! Despite seeing a medal on the public LB right now and (of course) hoping to not drop in the ranking, I'm aware that a drop by 1500 places on the private LB is something that should not be a complete shock given all the crowding just above 0.900.", "votes": null }, { "id": "1209325", "postDate": "02/18/2021 20:33:07", "content": "Same for me here, I wrote my own code in Tensorflow understanding every single line and this competition has been very instructive regardless of the final position.
\nConsidering that when the competition started I had no idea what a TFRecord was or how to use a TPU, now I feel enlightened and happy to see those weekly TPU hours finally being used!
\nUnfortunately, my CV and LB were not able to pass the 0.901 mark but it's ok, I'll go for the medal next time!😂
\nWish all of you the best!
Wow I've never thought about that, I also thought that a shake-up is very likely to happen but not in this way. Let's hope for the best for all of us!
", "rawMarkdown": "Wow I've never thought about that, I also thought that a shake-up is very likely to happen but not in this way. Let's hope for the best for all of us!", "votes": null } ], "comments": [ { "id": 1209265, "author_name": "bjoernholzhauer", "author_url": "", "post_date": "02/18/2021 19:31:13", "content": "You're within 0.004 of a silver medal on the public LB. If you end up in the medals, that would not be a shocking shake-up, at all - especially, if you took a very different approach of your own with a strong validation strategy - rather than relying on the public LB score or just using the same public notebook with a 0.903 score that many people appear to have used. On the other hand, despite having some models of my own that I think I optimized properly via CV, I've peaked out at what gets rounded to 0.903, too, so perhaps that turned out to be a natural limit for a lot of us?! Despite seeing a medal on the public LB right now and (of course) hoping to not drop in the ranking, I'm aware that a drop by 1500 places on the private LB is something that should not be a complete shock given all the crowding just above 0.900.
", "votes": null, "replies": [ { "id": 1209331, "author_name": "zowlex", "author_url": "", "post_date": "02/18/2021 20:36:11", "content": "Wow I've never thought about that, I also thought that a shake-up is very likely to happen but not in this way. Let's hope for the best for all of us!
", "votes": null, "replies": [] } ] }, { "id": 1209325, "author_name": "mviola", "author_url": "", "post_date": "02/18/2021 20:33:07", "content": "Same for me here, I wrote my own code in Tensorflow understanding every single line and this competition has been very instructive regardless of the final position.
\nConsidering that when the competition started I had no idea what a TFRecord was or how to use a TPU, now I feel enlightened and happy to see those weekly TPU hours finally being used!
\nUnfortunately, my CV and LB were not able to pass the 0.901 mark but it's ok, I'll go for the medal next time!😂
\nWish all of you the best!