I really don't want to make people feel bad with this post, but thought this was something amusing that I should share.
\n\nI wanted to participated in this competition a while back. To start out, being a fastai user, I started playing around with @iafoss's kernel. I submitted his kernel as well. I was planning to further explore this competition, but other projects took on a higher priority and I could not participate.
\n\nWhen the competition ended, I came to check out and see how my Kaggle friends and colleagues did in this competition and read some of the solutions. As I was scrolling through the leaderboard, I was in the bronze region when I was surprised to see my own name!
\n\nThat's right:
\n\nI was in shock: How was it that I put 5 minutes of effort into this competition and got a bronze?? While others spent 3 months working on this competition with a lower position then me, maybe not even getting a medal!
\n\nFor me, I think this is a testament to two things:
\n\nCongratulations to all the participants and thanks for sharing your solutions!
", "messages": [ { "id": "777716", "postDate": "03/17/2020 21:49:20", "content": "I really don't want to make people feel bad with this post, but thought this was something amusing that I should share.
\n\nI wanted to participated in this competition a while back. To start out, being a fastai user, I started playing around with @iafoss's kernel. I submitted his kernel as well. I was planning to further explore this competition, but other projects took on a higher priority and I could not participate.
\n\nWhen the competition ended, I came to check out and see how my Kaggle friends and colleagues did in this competition and read some of the solutions. As I was scrolling through the leaderboard, I was in the bronze region when I was surprised to see my own name!
\n\nThat's right:
\n\nI was in shock: How was it that I put 5 minutes of effort into this competition and got a bronze?? While others spent 3 months working on this competition with a lower position then me, maybe not even getting a medal!
\n\nFor me, I think this is a testament to two things:
\n\nCongratulations to all the participants and thanks for sharing your solutions!
", "rawMarkdown": "I really don't want to make people feel bad with this post, but thought this was something amusing that I should share.\n\n\nI wanted to participated in this competition a while back. To start out, being a fastai user, I started playing around with @iafoss's kernel. I submitted his [kernel](https://www.kaggle.com/iafoss/grapheme-fast-ai-starter-inference) as well. I was planning to further explore this competition, but other projects took on a higher priority and I could not participate. \n\nWhen the competition ended, I came to check out and see how my Kaggle friends and colleagues did in this competition and read some of the solutions. As I was scrolling through the leaderboard, I was in the bronze region when I was surprised to see my own name! \n\nThat's right: \n#_**I forked a public kernel, submitted it, and got bronze!**_\n\nI was in shock: How was it that I put 5 minutes of effort into this competition and got a bronze?? While others spent 3 months working on this competition with a lower position then me, maybe not even getting a medal! \n\nFor me, I think this is a testament to two things:\n\n1. The shakeup of this competition was clearly very rough, probably the worst in any Kaggle competition. Clearly, most contestants did not have the correct approach of developing a _generalizable_ model.\n2. Maybe medals aren't worth as much as I thought. They don't necessarily represent the amount of effort and skill put into a competition. Instead, I feel the sharing of knowledge here in the kernels and discussion forums are better representation of the effort and skill a Kaggler has. \n\nCongratulations to all the participants and thanks for sharing your solutions!", "votes": null }, { "id": "777739", "postDate": "03/17/2020 22:19:17", "content": "I know exactly where I made a wrong turn and have to completely own it.
\n\nI believe if it weren't for the fact that there were so many potentials for +10000 unseen graphemes in the test set, the shakeup wouldn't have happened. We were all told about this from the start. But when one is working on a solution and tries any and everything in their toolbox and only gets 0.95 LB (actually 0.93 private, but unknown to knew yet!!!!) then they get discouraged and start looking to the public techniques where millage may vary. The tricks that were shared on the forums and kernels were all legitimate computer vision techniques, it's just that people / we / I needed to not lose sight of the the real declared problem at hand / intent of the organizers, which was publicized from day 1.
", "rawMarkdown": "I know exactly where I made a wrong turn and have to completely own it.\n\nI believe if it weren't for the fact that there were so many potentials for +10000 unseen graphemes in the test set, the shakeup wouldn't have happened. We were all told about this from the start. But when one is working on a solution and tries any and everything in their toolbox and only gets 0.95 LB (actually 0.93 private, but unknown to knew yet!!!!) then they get discouraged and start looking to the public techniques where millage may vary. The tricks that were shared on the forums and kernels were all legitimate computer vision techniques, it's just that people / we / I needed to not lose sight of the the real declared problem at hand / intent of the organizers, which was publicized from day 1.", "votes": null }, { "id": "777956", "postDate": "03/18/2020 03:25:33", "content": "haha but you are making me feel bad😂 \nspent 2 months testing models but encountered a 200+ shake-down.\nhope for better luck next time
", "rawMarkdown": "haha but you are making me feel bad😂 \nspent 2 months testing models but encountered a 200+ shake-down.\nhope for better luck next time", "votes": null }, { "id": "778002", "postDate": "03/18/2020 04:24:38", "content": "Sorry this was not my intention. I wasn't sure if I should post this for this very reason. But I guess it just indicates that medals aren't everything but rather the knowledge and skills learned are more important.
", "rawMarkdown": "Sorry this was not my intention. I wasn't sure if I should post this for this very reason. But I guess it just indicates that medals aren't everything but rather the knowledge and skills learned are more important.", "votes": null }, { "id": "778022", "postDate": "03/18/2020 04:50:51", "content": "You are right!
", "rawMarkdown": "You are right!", "votes": null }, { "id": "778058", "postDate": "03/18/2020 05:42:10", "content": "@tanlikesmath That's fine. Thanks for sharing your thoughts. I think experiencing such shakes could be important for my future kaggling experience. Wish good luck for all of us!
", "rawMarkdown": "tanlikesmath That's fine. Thanks for sharing your thoughts. I think experiencing such shakes could be important for my future kaggling experience. Wish good luck for all of us!", "votes": null }, { "id": "778251", "postDate": "03/18/2020 09:22:15", "content": "Yeah, you are right..!! In any competition, it is always not about the medal it is actually about the amount of efforts that an individual puts in that competition. Spending his nights to run the model or doing analysis. For a beginnner like me, it doesn't matter whether I won or not or got any position. The thing is how much I learned new concepts or new methods to generate a more similar and effective effort. Thanks for the post. It motivates us
", "rawMarkdown": "Yeah, you are right..!! In any competition, it is always not about the medal it is actually about the amount of efforts that an individual puts in that competition. Spending his nights to run the model or doing analysis. For a beginnner like me, it doesn't matter whether I won or not or got any position. The thing is how much I learned new concepts or new methods to generate a more similar and effective effort. Thanks for the post. It motivates us", "votes": null }, { "id": "778259", "postDate": "03/18/2020 09:29:30", "content": "Good for you!
", "rawMarkdown": "Good for you!", "votes": null }, { "id": "778288", "postDate": "03/18/2020 10:11:47", "content": "About medals not mattering that much, I agree, in deep learning, there is some luck involved, and compute is also a bottleneck for many. Often in competitions where shakeup is big the bronze and silver medals may or may not be significant. But most of the time the gold medals are good solutions. That said, I feel one shouldn't chase the medals for anything, give your best effort, and keep on learning.
\n\nAlso, enjoy the high of competing (If you know what I mean) 😅
", "rawMarkdown": "About medals not mattering that much, I agree, in deep learning, there is some luck involved, and compute is also a bottleneck for many. Often in competitions where shakeup is big the bronze and silver medals may or may not be significant. But most of the time the gold medals are good solutions. That said, I feel one shouldn't chase the medals for anything, give your best effort, and keep on learning. \n\nAlso, enjoy the high of competing (If you know what I mean) 😅", "votes": null }, { "id": "778290", "postDate": "03/18/2020 10:15:27", "content": "About making models that don't generalize, I think this competition didn't have enough discussions about how important the unseen graphemes were, few sparse mentions here and there until 7 days ago when Qishen Ha posted his unseen validation scheme.
\n\nThe issue was that the CV and LB gap was always consistent, so its kind of an adversarial example to the normal cross-validation scheme and one had to do a deep dive into the data to find the reason, once you realize its a matter of finding ways to mitigate it, hence so many innovative solutions can be seen at the top.
", "rawMarkdown": "About making models that don't generalize, I think this competition didn't have enough discussions about how important the unseen graphemes were, few sparse mentions here and there until 7 days ago when Qishen Ha posted his unseen validation scheme. \n\nThe issue was that the CV and LB gap was always consistent, so its kind of an adversarial example to the normal cross-validation scheme and one had to do a deep dive into the data to find the reason, once you realize its a matter of finding ways to mitigate it, hence so many innovative solutions can be seen at the top.", "votes": null }, { "id": "778293", "postDate": "03/18/2020 10:19:15", "content": "@tonychenxyz Totally feel you man, even though I got lucky to realize about the unseen graphemes by the at the last moment and was able to weather the shakeup. I could have easily missed it. Would advise you to follow Heng's advise and write a solution post about what methodologies you followed and what you learnt, I tried it and it really helps collect your thoughts and keep you motivated.
", "rawMarkdown": "tonychenxyz Totally feel you man, even though I got lucky to realize about the unseen graphemes by the at the last moment and was able to weather the shakeup. I could have easily missed it. Would advise you to follow Heng's advise and write a solution post about what methodologies you followed and what you learnt, I tried it and it really helps collect your thoughts and keep you motivated.", "votes": null }, { "id": "778697", "postDate": "03/18/2020 16:44:08", "content": "Make it worth by investing more in the following competitions. Use any opportunity to learn and progress. And do not forgot to have fun in the process! Happy Kaggling!
", "rawMarkdown": "Make it worth by investing more in the following competitions. Use any opportunity to learn and progress. And do not forgot to have fun in the process! Happy Kaggling!", "votes": null }, { "id": "778899", "postDate": "03/18/2020 20:52:22", "content": "Thanks!
", "rawMarkdown": "Thanks!", "votes": null }, { "id": "778954", "postDate": "03/18/2020 22:09:26", "content": "Congratulations!!!!
", "rawMarkdown": "Congratulations!!!!", "votes": null } ], "comments": [ { "id": 777739, "author_name": "authman", "author_url": "", "post_date": "03/17/2020 22:19:17", "content": "I know exactly where I made a wrong turn and have to completely own it.
\n\nI believe if it weren't for the fact that there were so many potentials for +10000 unseen graphemes in the test set, the shakeup wouldn't have happened. We were all told about this from the start. But when one is working on a solution and tries any and everything in their toolbox and only gets 0.95 LB (actually 0.93 private, but unknown to knew yet!!!!) then they get discouraged and start looking to the public techniques where millage may vary. The tricks that were shared on the forums and kernels were all legitimate computer vision techniques, it's just that people / we / I needed to not lose sight of the the real declared problem at hand / intent of the organizers, which was publicized from day 1.
", "votes": null, "replies": [ { "id": 778022, "author_name": "kurianbenoy", "author_url": "", "post_date": "03/18/2020 04:50:51", "content": "You are right!
", "votes": null, "replies": [] } ] }, { "id": 777956, "author_name": "tonychenxyz", "author_url": "", "post_date": "03/18/2020 03:25:33", "content": "haha but you are making me feel bad😂 \nspent 2 months testing models but encountered a 200+ shake-down.\nhope for better luck next time
", "votes": null, "replies": [ { "id": 778002, "author_name": "tanlikesmath", "author_url": "", "post_date": "03/18/2020 04:24:38", "content": "Sorry this was not my intention. I wasn't sure if I should post this for this very reason. But I guess it just indicates that medals aren't everything but rather the knowledge and skills learned are more important.
", "votes": null, "replies": [] }, { "id": 778058, "author_name": "tonychenxyz", "author_url": "", "post_date": "03/18/2020 05:42:10", "content": "@tanlikesmath That's fine. Thanks for sharing your thoughts. I think experiencing such shakes could be important for my future kaggling experience. Wish good luck for all of us!
", "votes": null, "replies": [] }, { "id": 778293, "author_name": "dipamc77", "author_url": "", "post_date": "03/18/2020 10:19:15", "content": "@tonychenxyz Totally feel you man, even though I got lucky to realize about the unseen graphemes by the at the last moment and was able to weather the shakeup. I could have easily missed it. Would advise you to follow Heng's advise and write a solution post about what methodologies you followed and what you learnt, I tried it and it really helps collect your thoughts and keep you motivated.
", "votes": null, "replies": [] } ] }, { "id": 778251, "author_name": "usman4485", "author_url": "", "post_date": "03/18/2020 09:22:15", "content": "Yeah, you are right..!! In any competition, it is always not about the medal it is actually about the amount of efforts that an individual puts in that competition. Spending his nights to run the model or doing analysis. For a beginnner like me, it doesn't matter whether I won or not or got any position. The thing is how much I learned new concepts or new methods to generate a more similar and effective effort. Thanks for the post. It motivates us
", "votes": null, "replies": [] }, { "id": 778259, "author_name": "cpmpml", "author_url": "", "post_date": "03/18/2020 09:29:30", "content": "Good for you!
", "votes": null, "replies": [ { "id": 778899, "author_name": "tanlikesmath", "author_url": "", "post_date": "03/18/2020 20:52:22", "content": "Thanks!
", "votes": null, "replies": [] } ] }, { "id": 778288, "author_name": "dipamc77", "author_url": "", "post_date": "03/18/2020 10:11:47", "content": "About medals not mattering that much, I agree, in deep learning, there is some luck involved, and compute is also a bottleneck for many. Often in competitions where shakeup is big the bronze and silver medals may or may not be significant. But most of the time the gold medals are good solutions. That said, I feel one shouldn't chase the medals for anything, give your best effort, and keep on learning.
\n\nAlso, enjoy the high of competing (If you know what I mean) 😅
", "votes": null, "replies": [] }, { "id": 778290, "author_name": "dipamc77", "author_url": "", "post_date": "03/18/2020 10:15:27", "content": "About making models that don't generalize, I think this competition didn't have enough discussions about how important the unseen graphemes were, few sparse mentions here and there until 7 days ago when Qishen Ha posted his unseen validation scheme.
\n\nThe issue was that the CV and LB gap was always consistent, so its kind of an adversarial example to the normal cross-validation scheme and one had to do a deep dive into the data to find the reason, once you realize its a matter of finding ways to mitigate it, hence so many innovative solutions can be seen at the top.
", "votes": null, "replies": [] }, { "id": 778697, "author_name": "gpreda", "author_url": "", "post_date": "03/18/2020 16:44:08", "content": "Make it worth by investing more in the following competitions. Use any opportunity to learn and progress. And do not forgot to have fun in the process! Happy Kaggling!
", "votes": null, "replies": [] }, { "id": 778954, "author_name": "ferhatsahin", "author_url": "", "post_date": "03/18/2020 22:09:26", "content": "Congratulations!!!!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "777716": "I really don't want to make people feel bad with this post, but thought this was something amusing that I should share.\n\n\nI wanted to participated in this competition a while back. To start out, being a fastai user, I started playing around with @iafoss's kernel. I submitted his [kernel](https://www.kaggle.com/iafoss/grapheme-fast-ai-starter-inference) as well. I was planning to further explore this competition, but other projects took on a higher priority and I could not participate. \n\nWhen the competition ended, I came to check out and see how my Kaggle friends and colleagues did in this competition and read some of the solutions. As I was scrolling through the leaderboard, I was in the bronze region when I was surprised to see my own name! \n\nThat's right: \n#_**I forked a public kernel, submitted it, and got bronze!**_\n\nI was in shock: How was it that I put 5 minutes of effort into this competition and got a bronze?? While others spent 3 months working on this competition with a lower position then me, maybe not even getting a medal! \n\nFor me, I think this is a testament to two things:\n\n1. The shakeup of this competition was clearly very rough, probably the worst in any Kaggle competition. Clearly, most contestants did not have the correct approach of developing a _generalizable_ model.\n2. Maybe medals aren't worth as much as I thought. They don't necessarily represent the amount of effort and skill put into a competition. Instead, I feel the sharing of knowledge here in the kernels and discussion forums are better representation of the effort and skill a Kaggler has. \n\nCongratulations to all the participants and thanks for sharing your solutions!", "777739": "I know exactly where I made a wrong turn and have to completely own it.\n\nI believe if it weren't for the fact that there were so many potentials for +10000 unseen graphemes in the test set, the shakeup wouldn't have happened. We were all told about this from the start. But when one is working on a solution and tries any and everything in their toolbox and only gets 0.95 LB (actually 0.93 private, but unknown to knew yet!!!!) then they get discouraged and start looking to the public techniques where millage may vary. The tricks that were shared on the forums and kernels were all legitimate computer vision techniques, it's just that people / we / I needed to not lose sight of the the real declared problem at hand / intent of the organizers, which was publicized from day 1.", "777956": "haha but you are making me feel bad😂 \nspent 2 months testing models but encountered a 200+ shake-down.\nhope for better luck next time", "778002": "Sorry this was not my intention. I wasn't sure if I should post this for this very reason. But I guess it just indicates that medals aren't everything but rather the knowledge and skills learned are more important.", "778022": "You are right!", "778058": "tanlikesmath That's fine. Thanks for sharing your thoughts. I think experiencing such shakes could be important for my future kaggling experience. Wish good luck for all of us!", "778251": "Yeah, you are right..!! In any competition, it is always not about the medal it is actually about the amount of efforts that an individual puts in that competition. Spending his nights to run the model or doing analysis. For a beginnner like me, it doesn't matter whether I won or not or got any position. The thing is how much I learned new concepts or new methods to generate a more similar and effective effort. Thanks for the post. It motivates us", "778259": "Good for you!", "778288": "About medals not mattering that much, I agree, in deep learning, there is some luck involved, and compute is also a bottleneck for many. Often in competitions where shakeup is big the bronze and silver medals may or may not be significant. But most of the time the gold medals are good solutions. That said, I feel one shouldn't chase the medals for anything, give your best effort, and keep on learning. \n\nAlso, enjoy the high of competing (If you know what I mean) 😅", "778290": "About making models that don't generalize, I think this competition didn't have enough discussions about how important the unseen graphemes were, few sparse mentions here and there until 7 days ago when Qishen Ha posted his unseen validation scheme. \n\nThe issue was that the CV and LB gap was always consistent, so its kind of an adversarial example to the normal cross-validation scheme and one had to do a deep dive into the data to find the reason, once you realize its a matter of finding ways to mitigate it, hence so many innovative solutions can be seen at the top.", "778293": "tonychenxyz Totally feel you man, even though I got lucky to realize about the unseen graphemes by the at the last moment and was able to weather the shakeup. I could have easily missed it. Would advise you to follow Heng's advise and write a solution post about what methodologies you followed and what you learnt, I tried it and it really helps collect your thoughts and keep you motivated.", "778697": "Make it worth by investing more in the following competitions. Use any opportunity to learn and progress. And do not forgot to have fun in the process! Happy Kaggling!", "778899": "Thanks!", "778954": "Congratulations!!!!" }, "source": "meta" }