\n
\nThis doesn't much work here for me.
After recently joining Kaggle, I decided to take part in this competition. It was a wonderful experience. Here are a few realizations I've had as a result of this competition
\nKaggle is much more than running your model and making submissions. The major realization that I've had is that calling model.fit() is easy. Knowing which model to use and how to tweak it is the skill.
\n
\nThis doesn't much work here for me.
Kaggle is fun! I wouldn't have been doing it for a whole month if I didn't find it interesting. The learning opportunity and the satisfaction of seeing yourself improve over the course of the competition is unparallel.
\nYou see that - the public leaderboard and the high scoring public kernels - it's all a trap. I joined this competition with the goal to improve my computer vision knowledge and to learn. Call it a moment of weakness or frustration of seeing a succession of ideas not working, I fell in the trap of the high scoring kernels and the illusionary public leaderboard made me happy. Thank fully, everyone here was advising not to trust the public kernels and so I didn't include them in the final submissions.
\nHere's the catch. I trusted my CV, but turns out my CV was not trustable. I didn't even realize it until very late that my CV strategy was useless. Lessons learned. Note to self : Trust your CV only after you make sure(like 100% sure) that it is good.
\nOne of the most important task in any Kaggle competition is selecting the final 2/3 submissions. If done wrongly, you pay a very heavy price! So did I. Lost on a possible medal as my highest private scoring submission was not even a part of my final 3 submissions.
\n\n\n\"Mistakes made. Lessons learned.And much more to discover.\"
\n
Thank you everyone for this amazing learning opportunity. Every kernel and every discussion had something new to offer. Also, thanks a lot @cdeotte , you inspired, you helped and now I'm here to stay!
", "messages": [ { "id": 974682, "postDate": "2020-08-18T02:32:21.487Z", "content": "After recently joining Kaggle, I decided to take part in this competition. It was a wonderful experience. Here are a few realizations I've had as a result of this competition
\nKaggle is much more than running your model and making submissions. The major realization that I've had is that calling model.fit() is easy. Knowing which model to use and how to tweak it is the skill.
\n
\nThis doesn't much work here for me.
Kaggle is fun! I wouldn't have been doing it for a whole month if I didn't find it interesting. The learning opportunity and the satisfaction of seeing yourself improve over the course of the competition is unparallel.
\nYou see that - the public leaderboard and the high scoring public kernels - it's all a trap. I joined this competition with the goal to improve my computer vision knowledge and to learn. Call it a moment of weakness or frustration of seeing a succession of ideas not working, I fell in the trap of the high scoring kernels and the illusionary public leaderboard made me happy. Thank fully, everyone here was advising not to trust the public kernels and so I didn't include them in the final submissions.
\nHere's the catch. I trusted my CV, but turns out my CV was not trustable. I didn't even realize it until very late that my CV strategy was useless. Lessons learned. Note to self : Trust your CV only after you make sure(like 100% sure) that it is good.
\nOne of the most important task in any Kaggle competition is selecting the final 2/3 submissions. If done wrongly, you pay a very heavy price! So did I. Lost on a possible medal as my highest private scoring submission was not even a part of my final 3 submissions.
\n\n\n\"Mistakes made. Lessons learned.And much more to discover.\"
\n
Thank you everyone for this amazing learning opportunity. Every kernel and every discussion had something new to offer. Also, thanks a lot @cdeotte , you inspired, you helped and now I'm here to stay!
", "rawMarkdown": "After recently joining Kaggle, I decided to take part in this competition. It was a wonderful experience. Here are a few *realizations* I've had as a result of this competition\n\n## Kaggle Is Hard\nKaggle is much more than running your model and making submissions. The major realization that I've had is that calling model.fit() is easy. Knowing which model to use and how to tweak it is the skill. \n\nThis doesn't much work here for me.\n\n## But Kaggle is Fun\nKaggle is fun! I wouldn't have been doing it for a whole month if I didn't find it interesting. The learning opportunity and the satisfaction of seeing yourself improve over the course of the competition is unparallel. \n\n## The trap\nYou see that - the public leaderboard and the high scoring public kernels - it's all a trap. I joined this competition with the goal to improve my computer vision knowledge and to learn. Call it a moment of weakness or frustration of seeing a succession of ideas not working, I fell in the trap of the high scoring kernels and the illusionary public leaderboard made me happy. Thank fully, everyone here was advising not to trust the public kernels and so I didn't include them in the final submissions. \n\n## Trust your CV....eh\nHere's the catch. I trusted my CV, but turns out my CV was not trustable. I didn't even realize it until very late that my CV strategy was useless. Lessons learned. Note to self : Trust your CV only after you make sure(like 100% sure) that it is good.\n\n## The endgame\nOne of the most important task in any Kaggle competition is selecting the final 2/3 submissions. If done wrongly, you pay a very heavy price! So did I. Lost on a possible medal as my highest private scoring submission was not even a part of my final 3 submissions. \n\n> \"Mistakes made. Lessons learned.And much more to discover.\"\n\nThank you everyone for this amazing learning opportunity. Every kernel and every discussion had something new to offer. Also, thanks a lot @cdeotte , you inspired, you helped and now I'm here to stay!", "votes": 12 }, { "id": 974715, "postDate": "2020-08-18T03:02:01.313Z", "content": "I totally resonate with everything you've said Ayush! Even I had a really high CV. In fact our CV and LB were marginally apart. We expected the solution to be really stable. Huhhh. But, then shake-up happened
", "rawMarkdown": "I totally resonate with everything you've said Ayush! Even I had a really high CV. In fact our CV and LB were marginally apart. We expected the solution to be really stable. Huhhh. But, then shake-up happened", "votes": 1, "replies": [ { "id": 974782, "postDate": "2020-08-18T03:50:30.003Z", "content": "I was expecting a change in the final leaderboard. I unfortunately do not have prior experience to compare, but was this a drastic change in the leaderboard standings or this normal in other competitions as well?
", "rawMarkdown": "I was expecting a change in the final leaderboard. I unfortunately do not have prior experience to compare, but was this a drastic change in the leaderboard standings or this normal in other competitions as well?" }, { "id": 975400, "postDate": "2020-08-18T09:23:58.107Z", "content": "Same here man! this was my first competition, although throughout this competition I have observed the leaderboards of other competitions. They haven't shaken up that drastically (except the M5)
", "rawMarkdown": "Same here man! this was my first competition, although throughout this competition I have observed the leaderboards of other competitions. They haven't shaken up that drastically (except the M5) " } ] }, { "id": 991386, "postDate": "2020-08-30T11:25:04.353Z", "content": "I totally agree and have similar experiences. But I believe that it's more about the process than about the results! The more you learn by tuning your model, selecting different techniques, and iterating through different models, you involuntarily learn so much that you otherwise wouldn't have. And plus it's fun when it adds a competitive flavour and knowing that there are a whole bunch of data scientists probably doing things on lines similar to yours!
", "rawMarkdown": "I totally agree and have similar experiences. But I believe that it's more about the process than about the results! The more you learn by tuning your model, selecting different techniques, and iterating through different models, you involuntarily learn so much that you otherwise wouldn't have. And plus it's fun when it adds a competitive flavour and knowing that there are a whole bunch of data scientists probably doing things on lines similar to yours!", "votes": 2 }, { "id": 974698, "postDate": "2020-08-18T02:39:27.763Z", "content": "@foolofatook - These are great insights to have gained from your first competition. If you've learned all these things in just one competition then you will do great in the next one!
", "rawMarkdown": "@foolofatook - These are great insights to have gained from your first competition. If you've learned all these things in just one competition then you will do great in the next one!", "votes": 2, "replies": [ { "id": 974778, "postDate": "2020-08-18T03:46:43.580Z", "content": "Thanks a lot. This was, at least for me, the ideal first competition. It had everything. (Including drama. š)
", "rawMarkdown": "Thanks a lot. This was, at least for me, the ideal first competition. It had everything. (Including drama. š)", "votes": 2 }, { "id": 974841, "postDate": "2020-08-18T04:25:23.280Z", "content": "Literally had everything! This was my second \"real\" competition and I really agree with your comments.
", "rawMarkdown": "Literally had everything! This was my second \"real\" competition and I really agree with your comments. ", "votes": 1 }, { "id": 974842, "postDate": "2020-08-18T04:25:23.283Z", "rawMarkdown": "", "isDeleted": true }, { "id": 975242, "postDate": "2020-08-18T08:02:11.687Z", "content": "@santiviquez Hopefully, this learning experience transfers to other competitions.
", "rawMarkdown": "@santiviquez Hopefully, this learning experience transfers to other competitions. " }, { "id": 975857, "postDate": "2020-08-18T13:52:49.383Z", "content": "Hopefully yes, don't which one to start next. If the OSIC or the Cornel Birdcall one
", "rawMarkdown": "Hopefully yes, don't which one to start next. If the OSIC or the Cornel Birdcall one" } ] }, { "id": 991251, "postDate": "2020-08-30T09:11:20.280Z", "content": "So much to learn, so little time. Being on the same boat, I truly resonate with what you said !
", "rawMarkdown": "So much to learn, so little time. Being on the same boat, I truly resonate with what you said !" } ], "comments": [ { "id": 974715, "author_name": "Saransh Agarwal", "author_url": "", "post_date": "2020-08-18T03:02:01.313000", "content": "I totally resonate with everything you've said Ayush! Even I had a really high CV. In fact our CV and LB were marginally apart. We expected the solution to be really stable. Huhhh. But, then shake-up happened
", "votes": 1, "replies": [ { "id": 974782, "author_name": "Aayush Jain", "author_url": "", "post_date": "2020-08-18T03:50:30.003000", "content": "I was expecting a change in the final leaderboard. I unfortunately do not have prior experience to compare, but was this a drastic change in the leaderboard standings or this normal in other competitions as well?
", "votes": 0, "replies": [] }, { "id": 975400, "author_name": "Saransh Agarwal", "author_url": "", "post_date": "2020-08-18T09:23:58.107000", "content": "Same here man! this was my first competition, although throughout this competition I have observed the leaderboards of other competitions. They haven't shaken up that drastically (except the M5)
", "votes": 0, "replies": [] } ] }, { "id": 991386, "author_name": "Shyam R", "author_url": "", "post_date": "2020-08-30T11:25:04.353000", "content": "I totally agree and have similar experiences. But I believe that it's more about the process than about the results! The more you learn by tuning your model, selecting different techniques, and iterating through different models, you involuntarily learn so much that you otherwise wouldn't have. And plus it's fun when it adds a competitive flavour and knowing that there are a whole bunch of data scientists probably doing things on lines similar to yours!
", "votes": 2, "replies": [] }, { "id": 974698, "author_name": "Rob Mulla", "author_url": "", "post_date": "2020-08-18T02:39:27.763000", "content": "@foolofatook - These are great insights to have gained from your first competition. If you've learned all these things in just one competition then you will do great in the next one!
", "votes": 2, "replies": [ { "id": 974778, "author_name": "Aayush Jain", "author_url": "", "post_date": "2020-08-18T03:46:43.580000", "content": "Thanks a lot. This was, at least for me, the ideal first competition. It had everything. (Including drama. š)
", "votes": 2, "replies": [] }, { "id": 974841, "author_name": "Santiago Viquez", "author_url": "", "post_date": "2020-08-18T04:25:23.280000", "content": "Literally had everything! This was my second \"real\" competition and I really agree with your comments.
", "votes": 1, "replies": [] }, { "id": 974842, "author_name": "", "author_url": "", "post_date": "2020-08-18T04:25:23.283000", "content": "", "votes": 0, "replies": [] }, { "id": 975242, "author_name": "Aayush Jain", "author_url": "", "post_date": "2020-08-18T08:02:11.687000", "content": "@santiviquez Hopefully, this learning experience transfers to other competitions.
", "votes": 0, "replies": [] }, { "id": 975857, "author_name": "Santiago Viquez", "author_url": "", "post_date": "2020-08-18T13:52:49.383000", "content": "Hopefully yes, don't which one to start next. If the OSIC or the Cornel Birdcall one
", "votes": 0, "replies": [] } ] }, { "id": 991251, "author_name": "Akhil Upadhyaya", "author_url": "", "post_date": "2020-08-30T09:11:20.280000", "content": "So much to learn, so little time. Being on the same boat, I truly resonate with what you said !
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "974682": "After recently joining Kaggle, I decided to take part in this competition. It was a wonderful experience. Here are a few *realizations* I've had as a result of this competition\n\n## Kaggle Is Hard\nKaggle is much more than running your model and making submissions. The major realization that I've had is that calling model.fit() is easy. Knowing which model to use and how to tweak it is the skill. \n\nThis doesn't much work here for me.\n\n## But Kaggle is Fun\nKaggle is fun! I wouldn't have been doing it for a whole month if I didn't find it interesting. The learning opportunity and the satisfaction of seeing yourself improve over the course of the competition is unparallel. \n\n## The trap\nYou see that - the public leaderboard and the high scoring public kernels - it's all a trap. I joined this competition with the goal to improve my computer vision knowledge and to learn. Call it a moment of weakness or frustration of seeing a succession of ideas not working, I fell in the trap of the high scoring kernels and the illusionary public leaderboard made me happy. Thank fully, everyone here was advising not to trust the public kernels and so I didn't include them in the final submissions. \n\n## Trust your CV....eh\nHere's the catch. I trusted my CV, but turns out my CV was not trustable. I didn't even realize it until very late that my CV strategy was useless. Lessons learned. Note to self : Trust your CV only after you make sure(like 100% sure) that it is good.\n\n## The endgame\nOne of the most important task in any Kaggle competition is selecting the final 2/3 submissions. If done wrongly, you pay a very heavy price! So did I. Lost on a possible medal as my highest private scoring submission was not even a part of my final 3 submissions. \n\n> \"Mistakes made. Lessons learned.And much more to discover.\"\n\nThank you everyone for this amazing learning opportunity. Every kernel and every discussion had something new to offer. Also, thanks a lot @cdeotte , you inspired, you helped and now I'm here to stay!", "974715": "I totally resonate with everything you've said Ayush! Even I had a really high CV. In fact our CV and LB were marginally apart. We expected the solution to be really stable. Huhhh. But, then shake-up happened", "991386": "I totally agree and have similar experiences. But I believe that it's more about the process than about the results! The more you learn by tuning your model, selecting different techniques, and iterating through different models, you involuntarily learn so much that you otherwise wouldn't have. And plus it's fun when it adds a competitive flavour and knowing that there are a whole bunch of data scientists probably doing things on lines similar to yours!", "974698": "@foolofatook - These are great insights to have gained from your first competition. If you've learned all these things in just one competition then you will do great in the next one!", "991251": "So much to learn, so little time. Being on the same boat, I truly resonate with what you said !" } }