Here we are again, discussing late sharing issue. And for good reasons. It depends on what kind of kaggler one is, in my case it's only moderately annoying, here mostly to learn and enjoy. I almost never even bother to stack, or to use top public kernels, what of course means my code frequently doesn't even pass the cut of best public benchark, and that's ok for me.
\n\nBut for many others more than annoying it can be vital to have properly reflected their effort and ability in public leaderboard. And it is obvious this system flattens leaderboard meaning in a wrong way. It makes impossible to have our coding abilities benchmarked in a precise, or even true way.
\n\nMy point, though: How to change this? I mean, to some extent not only late kernels are problematic but any good shared kernel. Especially when someone can have no idea at all, zero, of machine learning and still push the csv submit button... and get a medal! If a 4 year old toddler can get a medal in Kaggle there is certainly a problem.
\n\nBut public kernels are such a good learning resource that I honestly don't see how to fix this without loosing one of Kaggle's best things. So, my question, anyone has a good idea? But, really, a good idea compatible with this premise of keeping kaggle a learning environment .
\n\nHere are my suggestions, only two patches but still something:
\n\n1) Remove possibility of submitting a kernel csv for 4 year old babies that can push a button. I mean...at least make it necessary to run the kernel, even if it is inside kernels environment.
\n\n2) Freeze posibility of submitting new kernel last say, 72 hours, maybe as much as one week before competition ending. (Unfreeze it from moment comp. ends, for solution sharing).
\n\nSo, any opinions on this patches, or any other better ones compatible with Kaggle \"way\"? It would be great not to be discussing/suffering this in every competition...
", "messages": [ { "id": 325448, "postDate": "2018-05-08T12:28:07.640Z", "content": "Here we are again, discussing late sharing issue. And for good reasons. It depends on what kind of kaggler one is, in my case it's only moderately annoying, here mostly to learn and enjoy. I almost never even bother to stack, or to use top public kernels, what of course means my code frequently doesn't even pass the cut of best public benchark, and that's ok for me.
\n\nBut for many others more than annoying it can be vital to have properly reflected their effort and ability in public leaderboard. And it is obvious this system flattens leaderboard meaning in a wrong way. It makes impossible to have our coding abilities benchmarked in a precise, or even true way.
\n\nMy point, though: How to change this? I mean, to some extent not only late kernels are problematic but any good shared kernel. Especially when someone can have no idea at all, zero, of machine learning and still push the csv submit button... and get a medal! If a 4 year old toddler can get a medal in Kaggle there is certainly a problem.
\n\nBut public kernels are such a good learning resource that I honestly don't see how to fix this without loosing one of Kaggle's best things. So, my question, anyone has a good idea? But, really, a good idea compatible with this premise of keeping kaggle a learning environment .
\n\nHere are my suggestions, only two patches but still something:
\n\n1) Remove possibility of submitting a kernel csv for 4 year old babies that can push a button. I mean...at least make it necessary to run the kernel, even if it is inside kernels environment.
\n\n2) Freeze posibility of submitting new kernel last say, 72 hours, maybe as much as one week before competition ending. (Unfreeze it from moment comp. ends, for solution sharing).
\n\nSo, any opinions on this patches, or any other better ones compatible with Kaggle \"way\"? It would be great not to be discussing/suffering this in every competition...
", "rawMarkdown": "**Here we are again**, discussing late sharing issue. And for good reasons. It depends on what kind of kaggler one is, in my case it's only moderately annoying, here mostly to learn and enjoy. I almost never even bother to stack, or to use top public kernels, what of course means my code frequently doesn't even pass the cut of best public benchark, and that's ok for me.\n\nBut for many others more than annoying it can be vital to have properly reflected their effort and ability in public leaderboard. And it is obvious this system flattens leaderboard meaning in a wrong way. **It makes impossible to have our coding abilities benchmarked in a precise, or even true way**. \n\nMy point, though: How to change this? I mean, to some extent not only late kernels are problematic but any good shared kernel. Especially when someone can have no idea at all, zero, of machine learning and still push the csv submit button... and get a medal! If a 4 year old toddler can get a medal in Kaggle there is certainly a problem. \n\n**But public kernels are such a good learning resource that I honestly don't see how to fix this without loosing one of Kaggle's best things**. So, my question, anyone has a good idea? But, really, a good idea compatible with this premise of keeping kaggle a learning environment .\n\nHere are my suggestions, only two patches but still something: \n\n1) Remove possibility of submitting a kernel csv for 4 year old babies that can push a button. I mean...at least make it necessary to run the kernel, even if it is inside kernels environment.\n\n2) Freeze posibility of submitting new kernel last say, 72 hours, maybe as much as one week before competition ending. (Unfreeze it from moment comp. ends, for solution sharing).\n\nSo, **any opinions on this patches, or any other better ones compatible with Kaggle \"way\"?** It would be great not to be discussing/suffering this in every competition...", "votes": 6 } ], "comments": [], "raw_markdown_by_id": { "325448": "**Here we are again**, discussing late sharing issue. And for good reasons. It depends on what kind of kaggler one is, in my case it's only moderately annoying, here mostly to learn and enjoy. I almost never even bother to stack, or to use top public kernels, what of course means my code frequently doesn't even pass the cut of best public benchark, and that's ok for me.\n\nBut for many others more than annoying it can be vital to have properly reflected their effort and ability in public leaderboard. And it is obvious this system flattens leaderboard meaning in a wrong way. **It makes impossible to have our coding abilities benchmarked in a precise, or even true way**. \n\nMy point, though: How to change this? I mean, to some extent not only late kernels are problematic but any good shared kernel. Especially when someone can have no idea at all, zero, of machine learning and still push the csv submit button... and get a medal! If a 4 year old toddler can get a medal in Kaggle there is certainly a problem. \n\n**But public kernels are such a good learning resource that I honestly don't see how to fix this without loosing one of Kaggle's best things**. So, my question, anyone has a good idea? But, really, a good idea compatible with this premise of keeping kaggle a learning environment .\n\nHere are my suggestions, only two patches but still something: \n\n1) Remove possibility of submitting a kernel csv for 4 year old babies that can push a button. I mean...at least make it necessary to run the kernel, even if it is inside kernels environment.\n\n2) Freeze posibility of submitting new kernel last say, 72 hours, maybe as much as one week before competition ending. (Unfreeze it from moment comp. ends, for solution sharing).\n\nSo, **any opinions on this patches, or any other better ones compatible with Kaggle \"way\"?** It would be great not to be discussing/suffering this in every competition..." } }